Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 1 | #include <linux/ceph/ceph_debug.h> |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 2 | |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 3 | #include <linux/module.h> |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 4 | #include <linux/sched.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 5 | #include <linux/slab.h> |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 6 | #include <linux/file.h> |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 7 | #include <linux/mount.h> |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 8 | #include <linux/namei.h> |
| 9 | #include <linux/writeback.h> |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 10 | #include <linux/falloc.h> |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 11 | |
| 12 | #include "super.h" |
| 13 | #include "mds_client.h" |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 14 | #include "cache.h" |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 15 | |
| 16 | /* |
| 17 | * Ceph file operations |
| 18 | * |
| 19 | * Implement basic open/close functionality, and implement |
| 20 | * read/write. |
| 21 | * |
| 22 | * We implement three modes of file I/O: |
| 23 | * - buffered uses the generic_file_aio_{read,write} helpers |
| 24 | * |
| 25 | * - synchronous is used when there is multi-client read/write |
| 26 | * sharing, avoids the page cache, and synchronously waits for an |
| 27 | * ack from the OSD. |
| 28 | * |
| 29 | * - direct io takes the variant of the sync path that references |
| 30 | * user pages directly. |
| 31 | * |
| 32 | * fsync() flushes and waits on dirty pages, but just queues metadata |
| 33 | * for writeback: since the MDS can recover size and mtime there is no |
| 34 | * need to wait for MDS acknowledgement. |
| 35 | */ |
| 36 | |
Zhu, Caifeng | b5b98989d | 2015-10-08 15:26:15 +0800 | [diff] [blame] | 37 | /* |
| 38 | * Calculate the length sum of direct io vectors that can |
| 39 | * be combined into one page vector. |
| 40 | */ |
| 41 | static size_t dio_get_pagev_size(const struct iov_iter *it) |
| 42 | { |
| 43 | const struct iovec *iov = it->iov; |
| 44 | const struct iovec *iovend = iov + it->nr_segs; |
| 45 | size_t size; |
| 46 | |
| 47 | size = iov->iov_len - it->iov_offset; |
| 48 | /* |
| 49 | * An iov can be page vectored when both the current tail |
| 50 | * and the next base are page aligned. |
| 51 | */ |
| 52 | while (PAGE_ALIGNED((iov->iov_base + iov->iov_len)) && |
| 53 | (++iov < iovend && PAGE_ALIGNED((iov->iov_base)))) { |
| 54 | size += iov->iov_len; |
| 55 | } |
| 56 | dout("dio_get_pagevlen len = %zu\n", size); |
| 57 | return size; |
| 58 | } |
| 59 | |
| 60 | /* |
| 61 | * Allocate a page vector based on (@it, @nbytes). |
| 62 | * The return value is the tuple describing a page vector, |
| 63 | * that is (@pages, @page_align, @num_pages). |
| 64 | */ |
| 65 | static struct page ** |
| 66 | dio_get_pages_alloc(const struct iov_iter *it, size_t nbytes, |
| 67 | size_t *page_align, int *num_pages) |
| 68 | { |
| 69 | struct iov_iter tmp_it = *it; |
| 70 | size_t align; |
| 71 | struct page **pages; |
| 72 | int ret = 0, idx, npages; |
| 73 | |
| 74 | align = (unsigned long)(it->iov->iov_base + it->iov_offset) & |
| 75 | (PAGE_SIZE - 1); |
| 76 | npages = calc_pages_for(align, nbytes); |
| 77 | pages = kmalloc(sizeof(*pages) * npages, GFP_KERNEL); |
| 78 | if (!pages) { |
| 79 | pages = vmalloc(sizeof(*pages) * npages); |
| 80 | if (!pages) |
| 81 | return ERR_PTR(-ENOMEM); |
| 82 | } |
| 83 | |
| 84 | for (idx = 0; idx < npages; ) { |
| 85 | size_t start; |
| 86 | ret = iov_iter_get_pages(&tmp_it, pages + idx, nbytes, |
| 87 | npages - idx, &start); |
| 88 | if (ret < 0) |
| 89 | goto fail; |
| 90 | |
| 91 | iov_iter_advance(&tmp_it, ret); |
| 92 | nbytes -= ret; |
| 93 | idx += (ret + start + PAGE_SIZE - 1) / PAGE_SIZE; |
| 94 | } |
| 95 | |
| 96 | BUG_ON(nbytes != 0); |
| 97 | *num_pages = npages; |
| 98 | *page_align = align; |
| 99 | dout("dio_get_pages_alloc: got %d pages align %zu\n", npages, align); |
| 100 | return pages; |
| 101 | fail: |
| 102 | ceph_put_page_vector(pages, idx, false); |
| 103 | return ERR_PTR(ret); |
| 104 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 105 | |
| 106 | /* |
| 107 | * Prepare an open request. Preallocate ceph_cap to avoid an |
| 108 | * inopportune ENOMEM later. |
| 109 | */ |
| 110 | static struct ceph_mds_request * |
| 111 | prepare_open_request(struct super_block *sb, int flags, int create_mode) |
| 112 | { |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 113 | struct ceph_fs_client *fsc = ceph_sb_to_client(sb); |
| 114 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 115 | struct ceph_mds_request *req; |
| 116 | int want_auth = USE_ANY_MDS; |
| 117 | int op = (flags & O_CREAT) ? CEPH_MDS_OP_CREATE : CEPH_MDS_OP_OPEN; |
| 118 | |
| 119 | if (flags & (O_WRONLY|O_RDWR|O_CREAT|O_TRUNC)) |
| 120 | want_auth = USE_AUTH_MDS; |
| 121 | |
| 122 | req = ceph_mdsc_create_request(mdsc, op, want_auth); |
| 123 | if (IS_ERR(req)) |
| 124 | goto out; |
| 125 | req->r_fmode = ceph_flags_to_mode(flags); |
| 126 | req->r_args.open.flags = cpu_to_le32(flags); |
| 127 | req->r_args.open.mode = cpu_to_le32(create_mode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 128 | out: |
| 129 | return req; |
| 130 | } |
| 131 | |
| 132 | /* |
| 133 | * initialize private struct file data. |
| 134 | * if we fail, clean up by dropping fmode reference on the ceph_inode |
| 135 | */ |
| 136 | static int ceph_init_file(struct inode *inode, struct file *file, int fmode) |
| 137 | { |
| 138 | struct ceph_file_info *cf; |
| 139 | int ret = 0; |
| 140 | |
| 141 | switch (inode->i_mode & S_IFMT) { |
| 142 | case S_IFREG: |
Yan, Zheng | 46b59b2 | 2016-05-18 15:25:03 +0800 | [diff] [blame] | 143 | ceph_fscache_register_inode_cookie(inode); |
| 144 | ceph_fscache_file_set_cookie(inode, file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 145 | case S_IFDIR: |
| 146 | dout("init_file %p %p 0%o (regular)\n", inode, file, |
| 147 | inode->i_mode); |
Geliang Tang | 99ec269 | 2016-03-13 15:26:29 +0800 | [diff] [blame] | 148 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 149 | if (cf == NULL) { |
| 150 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 151 | return -ENOMEM; |
| 152 | } |
| 153 | cf->fmode = fmode; |
| 154 | cf->next_offset = 2; |
Yan, Zheng | fdd4e15 | 2015-06-16 20:48:56 +0800 | [diff] [blame] | 155 | cf->readdir_cache_idx = -1; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 156 | file->private_data = cf; |
| 157 | BUG_ON(inode->i_fop->release != ceph_release); |
| 158 | break; |
| 159 | |
| 160 | case S_IFLNK: |
| 161 | dout("init_file %p %p 0%o (symlink)\n", inode, file, |
| 162 | inode->i_mode); |
| 163 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 164 | break; |
| 165 | |
| 166 | default: |
| 167 | dout("init_file %p %p 0%o (special)\n", inode, file, |
| 168 | inode->i_mode); |
| 169 | /* |
| 170 | * we need to drop the open ref now, since we don't |
| 171 | * have .release set to ceph_release. |
| 172 | */ |
| 173 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 174 | BUG_ON(inode->i_fop->release == ceph_release); |
| 175 | |
| 176 | /* call the proper open fop */ |
| 177 | ret = inode->i_fop->open(inode, file); |
| 178 | } |
| 179 | return ret; |
| 180 | } |
| 181 | |
| 182 | /* |
Yan, Zheng | 7731032 | 2016-04-08 15:27:16 +0800 | [diff] [blame] | 183 | * try renew caps after session gets killed. |
| 184 | */ |
| 185 | int ceph_renew_caps(struct inode *inode) |
| 186 | { |
| 187 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
| 188 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 189 | struct ceph_mds_request *req; |
| 190 | int err, flags, wanted; |
| 191 | |
| 192 | spin_lock(&ci->i_ceph_lock); |
| 193 | wanted = __ceph_caps_file_wanted(ci); |
| 194 | if (__ceph_is_any_real_caps(ci) && |
| 195 | (!(wanted & CEPH_CAP_ANY_WR) == 0 || ci->i_auth_cap)) { |
| 196 | int issued = __ceph_caps_issued(ci, NULL); |
| 197 | spin_unlock(&ci->i_ceph_lock); |
| 198 | dout("renew caps %p want %s issued %s updating mds_wanted\n", |
| 199 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); |
| 200 | ceph_check_caps(ci, 0, NULL); |
| 201 | return 0; |
| 202 | } |
| 203 | spin_unlock(&ci->i_ceph_lock); |
| 204 | |
| 205 | flags = 0; |
| 206 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) |
| 207 | flags = O_RDWR; |
| 208 | else if (wanted & CEPH_CAP_FILE_RD) |
| 209 | flags = O_RDONLY; |
| 210 | else if (wanted & CEPH_CAP_FILE_WR) |
| 211 | flags = O_WRONLY; |
| 212 | #ifdef O_LAZY |
| 213 | if (wanted & CEPH_CAP_FILE_LAZYIO) |
| 214 | flags |= O_LAZY; |
| 215 | #endif |
| 216 | |
| 217 | req = prepare_open_request(inode->i_sb, flags, 0); |
| 218 | if (IS_ERR(req)) { |
| 219 | err = PTR_ERR(req); |
| 220 | goto out; |
| 221 | } |
| 222 | |
| 223 | req->r_inode = inode; |
| 224 | ihold(inode); |
| 225 | req->r_num_caps = 1; |
| 226 | req->r_fmode = -1; |
| 227 | |
| 228 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
| 229 | ceph_mdsc_put_request(req); |
| 230 | out: |
| 231 | dout("renew caps %p open result=%d\n", inode, err); |
| 232 | return err < 0 ? err : 0; |
| 233 | } |
| 234 | |
| 235 | /* |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 236 | * If we already have the requisite capabilities, we can satisfy |
| 237 | * the open request locally (no need to request new caps from the |
| 238 | * MDS). We do, however, need to inform the MDS (asynchronously) |
| 239 | * if our wanted caps set expands. |
| 240 | */ |
| 241 | int ceph_open(struct inode *inode, struct file *file) |
| 242 | { |
| 243 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 244 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
| 245 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 246 | struct ceph_mds_request *req; |
| 247 | struct ceph_file_info *cf = file->private_data; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 248 | int err; |
| 249 | int flags, fmode, wanted; |
| 250 | |
| 251 | if (cf) { |
| 252 | dout("open file %p is already opened\n", file); |
| 253 | return 0; |
| 254 | } |
| 255 | |
| 256 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ |
| 257 | flags = file->f_flags & ~(O_CREAT|O_EXCL); |
| 258 | if (S_ISDIR(inode->i_mode)) |
| 259 | flags = O_DIRECTORY; /* mds likes to know */ |
| 260 | |
| 261 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, |
| 262 | ceph_vinop(inode), file, flags, file->f_flags); |
| 263 | fmode = ceph_flags_to_mode(flags); |
| 264 | wanted = ceph_caps_for_mode(fmode); |
| 265 | |
| 266 | /* snapped files are read-only */ |
| 267 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) |
| 268 | return -EROFS; |
| 269 | |
| 270 | /* trivially open snapdir */ |
| 271 | if (ceph_snap(inode) == CEPH_SNAPDIR) { |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 272 | spin_lock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 273 | __ceph_get_fmode(ci, fmode); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 274 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 275 | return ceph_init_file(inode, file, fmode); |
| 276 | } |
| 277 | |
| 278 | /* |
Sage Weil | 7421ab8 | 2010-11-07 09:07:15 -0800 | [diff] [blame] | 279 | * No need to block if we have caps on the auth MDS (for |
| 280 | * write) or any MDS (for read). Update wanted set |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 281 | * asynchronously. |
| 282 | */ |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 283 | spin_lock(&ci->i_ceph_lock); |
Sage Weil | 7421ab8 | 2010-11-07 09:07:15 -0800 | [diff] [blame] | 284 | if (__ceph_is_any_real_caps(ci) && |
| 285 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 286 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
| 287 | int issued = __ceph_caps_issued(ci, NULL); |
| 288 | |
| 289 | dout("open %p fmode %d want %s issued %s using existing\n", |
| 290 | inode, fmode, ceph_cap_string(wanted), |
| 291 | ceph_cap_string(issued)); |
| 292 | __ceph_get_fmode(ci, fmode); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 293 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 294 | |
| 295 | /* adjust wanted? */ |
| 296 | if ((issued & wanted) != wanted && |
| 297 | (mds_wanted & wanted) != wanted && |
| 298 | ceph_snap(inode) != CEPH_SNAPDIR) |
| 299 | ceph_check_caps(ci, 0, NULL); |
| 300 | |
| 301 | return ceph_init_file(inode, file, fmode); |
| 302 | } else if (ceph_snap(inode) != CEPH_NOSNAP && |
| 303 | (ci->i_snap_caps & wanted) == wanted) { |
| 304 | __ceph_get_fmode(ci, fmode); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 305 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 306 | return ceph_init_file(inode, file, fmode); |
| 307 | } |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 308 | |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 309 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 310 | |
| 311 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); |
| 312 | req = prepare_open_request(inode->i_sb, flags, 0); |
| 313 | if (IS_ERR(req)) { |
| 314 | err = PTR_ERR(req); |
| 315 | goto out; |
| 316 | } |
Sage Weil | 70b666c | 2011-05-27 09:24:26 -0700 | [diff] [blame] | 317 | req->r_inode = inode; |
| 318 | ihold(inode); |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 319 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 320 | req->r_num_caps = 1; |
Jianpeng Ma | e36d571 | 2015-08-18 10:25:35 +0800 | [diff] [blame] | 321 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 322 | if (!err) |
| 323 | err = ceph_init_file(inode, file, req->r_fmode); |
| 324 | ceph_mdsc_put_request(req); |
| 325 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); |
| 326 | out: |
| 327 | return err; |
| 328 | } |
| 329 | |
| 330 | |
| 331 | /* |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 332 | * Do a lookup + open with a single request. If we get a non-existent |
| 333 | * file or symlink, return 1 so the VFS can retry. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 334 | */ |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 335 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
Al Viro | 30d9049 | 2012-06-22 12:40:19 +0400 | [diff] [blame] | 336 | struct file *file, unsigned flags, umode_t mode, |
Al Viro | d958527 | 2012-06-22 12:39:14 +0400 | [diff] [blame] | 337 | int *opened) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 338 | { |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 339 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
| 340 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 341 | struct ceph_mds_request *req; |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 342 | struct dentry *dn; |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 343 | struct ceph_acls_info acls = {}; |
Yan, Zheng | 315f240 | 2016-03-07 10:34:50 +0800 | [diff] [blame] | 344 | int mask; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 345 | int err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 346 | |
Al Viro | a455589 | 2014-10-21 20:11:25 -0400 | [diff] [blame] | 347 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
| 348 | dir, dentry, dentry, |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 349 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
| 350 | |
| 351 | if (dentry->d_name.len > NAME_MAX) |
| 352 | return -ENAMETOOLONG; |
| 353 | |
| 354 | err = ceph_init_dentry(dentry); |
| 355 | if (err < 0) |
| 356 | return err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 357 | |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 358 | if (flags & O_CREAT) { |
| 359 | err = ceph_pre_init_acls(dir, &mode, &acls); |
| 360 | if (err < 0) |
| 361 | return err; |
| 362 | } |
| 363 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 364 | /* do the open */ |
| 365 | req = prepare_open_request(dir->i_sb, flags, mode); |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 366 | if (IS_ERR(req)) { |
| 367 | err = PTR_ERR(req); |
| 368 | goto out_acl; |
| 369 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 370 | req->r_dentry = dget(dentry); |
| 371 | req->r_num_caps = 2; |
| 372 | if (flags & O_CREAT) { |
| 373 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; |
| 374 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 375 | if (acls.pagelist) { |
| 376 | req->r_pagelist = acls.pagelist; |
| 377 | acls.pagelist = NULL; |
| 378 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 379 | } |
Yan, Zheng | 315f240 | 2016-03-07 10:34:50 +0800 | [diff] [blame] | 380 | |
| 381 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; |
| 382 | if (ceph_security_xattr_wanted(dir)) |
| 383 | mask |= CEPH_CAP_XATTR_SHARED; |
| 384 | req->r_args.open.mask = cpu_to_le32(mask); |
| 385 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 386 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ |
Sage Weil | acda765 | 2011-07-26 11:27:48 -0700 | [diff] [blame] | 387 | err = ceph_mdsc_do_request(mdsc, |
| 388 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, |
| 389 | req); |
Yan, Zheng | bf91c31 | 2015-01-19 13:23:20 +0800 | [diff] [blame] | 390 | err = ceph_handle_snapdir(req, dentry, err); |
Sam Lang | 79aec984 | 2012-12-19 09:44:23 -1000 | [diff] [blame] | 391 | if (err) |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 392 | goto out_req; |
Sam Lang | 79aec984 | 2012-12-19 09:44:23 -1000 | [diff] [blame] | 393 | |
Jianpeng Ma | a43137f | 2015-08-18 10:23:50 +0800 | [diff] [blame] | 394 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 395 | err = ceph_handle_notrace_create(dir, dentry); |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 396 | |
Al Viro | 00699ad | 2016-07-05 09:44:53 -0400 | [diff] [blame] | 397 | if (d_in_lookup(dentry)) { |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 398 | dn = ceph_finish_lookup(req, dentry, err); |
| 399 | if (IS_ERR(dn)) |
| 400 | err = PTR_ERR(dn); |
| 401 | } else { |
| 402 | /* we were given a hashed negative dentry */ |
| 403 | dn = NULL; |
| 404 | } |
Sage Weil | 468640e | 2011-07-26 11:28:11 -0700 | [diff] [blame] | 405 | if (err) |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 406 | goto out_req; |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 407 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 408 | /* make vfs retry on splice, ENOENT, or symlink */ |
| 409 | dout("atomic_open finish_no_open on dn %p\n", dn); |
| 410 | err = finish_no_open(file, dn); |
| 411 | } else { |
| 412 | dout("atomic_open finish_open on dn %p\n", dn); |
Sam Lang | 6e8575f | 2012-12-28 09:56:46 -0800 | [diff] [blame] | 413 | if (req->r_op == CEPH_MDS_OP_CREATE && req->r_reply_info.has_create_ino) { |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 414 | ceph_init_inode_acls(d_inode(dentry), &acls); |
Sam Lang | 6e8575f | 2012-12-28 09:56:46 -0800 | [diff] [blame] | 415 | *opened |= FILE_CREATED; |
| 416 | } |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 417 | err = finish_open(file, dentry, ceph_open, opened); |
| 418 | } |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 419 | out_req: |
Yan, Zheng | ab86654 | 2014-04-01 20:20:17 +0800 | [diff] [blame] | 420 | if (!req->r_err && req->r_target_inode) |
| 421 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 422 | ceph_mdsc_put_request(req); |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 423 | out_acl: |
| 424 | ceph_release_acls_info(&acls); |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 425 | dout("atomic_open result=%d\n", err); |
Al Viro | d958527 | 2012-06-22 12:39:14 +0400 | [diff] [blame] | 426 | return err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 427 | } |
| 428 | |
| 429 | int ceph_release(struct inode *inode, struct file *file) |
| 430 | { |
| 431 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 432 | struct ceph_file_info *cf = file->private_data; |
| 433 | |
| 434 | dout("release inode %p file %p\n", inode, file); |
| 435 | ceph_put_fmode(ci, cf->fmode); |
| 436 | if (cf->last_readdir) |
| 437 | ceph_mdsc_put_request(cf->last_readdir); |
| 438 | kfree(cf->last_name); |
| 439 | kfree(cf->dir_info); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 440 | kmem_cache_free(ceph_file_cachep, cf); |
Sage Weil | 195d3ce | 2010-03-01 09:57:54 -0800 | [diff] [blame] | 441 | |
| 442 | /* wake up anyone waiting for caps on this inode */ |
Yehuda Sadeh | 03066f2 | 2010-07-27 13:11:08 -0700 | [diff] [blame] | 443 | wake_up_all(&ci->i_cap_wq); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 444 | return 0; |
| 445 | } |
| 446 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 447 | enum { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 448 | HAVE_RETRIED = 1, |
| 449 | CHECK_EOF = 2, |
| 450 | READ_INLINE = 3, |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 451 | }; |
| 452 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 453 | /* |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 454 | * Read a range of bytes striped over one or more objects. Iterate over |
| 455 | * objects we stripe over. (That's not atomic, but good enough for now.) |
| 456 | * |
| 457 | * If we get a short result from the OSD, check against i_size; we need to |
| 458 | * only return a short read to the caller if we hit EOF. |
| 459 | */ |
| 460 | static int striped_read(struct inode *inode, |
| 461 | u64 off, u64 len, |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 462 | struct page **pages, int num_pages, |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 463 | int *checkeof) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 464 | { |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 465 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 466 | struct ceph_inode_info *ci = ceph_inode(inode); |
Dan Carpenter | 688bac4 | 2013-07-23 16:48:01 +0300 | [diff] [blame] | 467 | u64 pos, this_len, left; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 468 | loff_t i_size; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 469 | int page_align, pages_left; |
| 470 | int read, ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 471 | struct page **page_pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 472 | bool hit_stripe, was_short; |
| 473 | |
| 474 | /* |
| 475 | * we may need to do multiple reads. not atomic, unfortunately. |
| 476 | */ |
| 477 | pos = off; |
| 478 | left = len; |
| 479 | page_pos = pages; |
| 480 | pages_left = num_pages; |
| 481 | read = 0; |
| 482 | |
| 483 | more: |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 484 | page_align = pos & ~PAGE_MASK; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 485 | this_len = left; |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 486 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 487 | &ci->i_layout, pos, &this_len, |
| 488 | ci->i_truncate_seq, |
| 489 | ci->i_truncate_size, |
Sage Weil | b7495fc | 2010-11-09 12:43:12 -0800 | [diff] [blame] | 490 | page_pos, pages_left, page_align); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 491 | if (ret == -ENOENT) |
| 492 | ret = 0; |
Sage Weil | 0e98728 | 2011-06-07 20:40:35 -0700 | [diff] [blame] | 493 | hit_stripe = this_len < left; |
| 494 | was_short = ret >= 0 && ret < this_len; |
Dan Carpenter | 688bac4 | 2013-07-23 16:48:01 +0300 | [diff] [blame] | 495 | dout("striped_read %llu~%llu (read %u) got %d%s%s\n", pos, left, read, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 496 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
| 497 | |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 498 | i_size = i_size_read(inode); |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 499 | if (ret >= 0) { |
| 500 | int didpages; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 501 | if (was_short && (pos + ret < i_size)) { |
| 502 | int zlen = min(this_len - ret, i_size - pos - ret); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 503 | int zoff = (off & ~PAGE_MASK) + read + ret; |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 504 | dout(" zero gap %llu to %llu\n", |
Yan, Zheng | 1487a68 | 2015-01-06 15:29:14 +0800 | [diff] [blame] | 505 | pos + ret, pos + ret + zlen); |
| 506 | ceph_zero_page_vector_range(zoff, zlen, pages); |
| 507 | ret += zlen; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 508 | } |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 509 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 510 | didpages = (page_align + ret) >> PAGE_SHIFT; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 511 | pos += ret; |
| 512 | read = pos - off; |
| 513 | left -= ret; |
| 514 | page_pos += didpages; |
| 515 | pages_left -= didpages; |
| 516 | |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 517 | /* hit stripe and need continue*/ |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 518 | if (left && hit_stripe && pos < i_size) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 519 | goto more; |
| 520 | } |
| 521 | |
majianpeng | ee7289b | 2013-08-21 15:02:51 +0800 | [diff] [blame] | 522 | if (read > 0) { |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 523 | ret = read; |
Sage Weil | c3cd628 | 2011-06-01 16:08:44 -0700 | [diff] [blame] | 524 | /* did we bounce off eof? */ |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 525 | if (pos + left > i_size) |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 526 | *checkeof = CHECK_EOF; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 527 | } |
| 528 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 529 | dout("striped_read returns %d\n", ret); |
| 530 | return ret; |
| 531 | } |
| 532 | |
| 533 | /* |
| 534 | * Completely synchronous read and write methods. Direct from __user |
| 535 | * buffer to osd, or directly to user pages (if O_DIRECT). |
| 536 | * |
| 537 | * If the read spans object boundary, just do multiple reads. |
| 538 | */ |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 539 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
| 540 | int *checkeof) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 541 | { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 542 | struct file *file = iocb->ki_filp; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 543 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 544 | struct page **pages; |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 545 | u64 off = iocb->ki_pos; |
Henry C Chang | ab226e2 | 2010-12-15 20:41:54 -0800 | [diff] [blame] | 546 | int num_pages, ret; |
Al Viro | 2b777c9 | 2014-04-03 22:31:22 -0400 | [diff] [blame] | 547 | size_t len = iov_iter_count(i); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 548 | |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 549 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
| 550 | (unsigned)len, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 551 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
Yan, Zheng | d0d0db2 | 2014-07-21 10:15:48 +0800 | [diff] [blame] | 552 | |
| 553 | if (!len) |
| 554 | return 0; |
Sage Weil | e98b6fe | 2010-11-09 12:24:53 -0800 | [diff] [blame] | 555 | /* |
| 556 | * flush any page cache pages in this range. this |
| 557 | * will make concurrent normal and sync io slow, |
| 558 | * but it will at least behave sensibly when they are |
| 559 | * in sequence. |
| 560 | */ |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 561 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
| 562 | off + len); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 563 | if (ret < 0) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 564 | return ret; |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 565 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 566 | num_pages = calc_pages_for(off, len); |
| 567 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
| 568 | if (IS_ERR(pages)) |
| 569 | return PTR_ERR(pages); |
| 570 | ret = striped_read(inode, off, len, pages, |
| 571 | num_pages, checkeof); |
| 572 | if (ret > 0) { |
| 573 | int l, k = 0; |
| 574 | size_t left = ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 575 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 576 | while (left) { |
| 577 | size_t page_off = off & ~PAGE_MASK; |
| 578 | size_t copy = min_t(size_t, left, |
| 579 | PAGE_SIZE - page_off); |
| 580 | l = copy_page_to_iter(pages[k++], page_off, copy, i); |
| 581 | off += l; |
| 582 | left -= l; |
| 583 | if (l < copy) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 584 | break; |
| 585 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 586 | } |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 587 | ceph_release_page_vector(pages, num_pages); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 588 | |
| 589 | if (off > iocb->ki_pos) { |
| 590 | ret = off - iocb->ki_pos; |
| 591 | iocb->ki_pos = off; |
| 592 | } |
| 593 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 594 | dout("sync_read result %d\n", ret); |
| 595 | return ret; |
| 596 | } |
| 597 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 598 | struct ceph_aio_request { |
| 599 | struct kiocb *iocb; |
| 600 | size_t total_len; |
| 601 | int write; |
| 602 | int error; |
| 603 | struct list_head osd_reqs; |
| 604 | unsigned num_reqs; |
| 605 | atomic_t pending_reqs; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 606 | struct timespec mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 607 | struct ceph_cap_flush *prealloc_cf; |
| 608 | }; |
| 609 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 610 | struct ceph_aio_work { |
| 611 | struct work_struct work; |
| 612 | struct ceph_osd_request *req; |
| 613 | }; |
| 614 | |
| 615 | static void ceph_aio_retry_work(struct work_struct *work); |
| 616 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 617 | static void ceph_aio_complete(struct inode *inode, |
| 618 | struct ceph_aio_request *aio_req) |
| 619 | { |
| 620 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 621 | int ret; |
| 622 | |
| 623 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) |
| 624 | return; |
| 625 | |
| 626 | ret = aio_req->error; |
| 627 | if (!ret) |
| 628 | ret = aio_req->total_len; |
| 629 | |
| 630 | dout("ceph_aio_complete %p rc %d\n", inode, ret); |
| 631 | |
| 632 | if (ret >= 0 && aio_req->write) { |
| 633 | int dirty; |
| 634 | |
| 635 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; |
| 636 | if (endoff > i_size_read(inode)) { |
| 637 | if (ceph_inode_set_size(inode, endoff)) |
| 638 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); |
| 639 | } |
| 640 | |
| 641 | spin_lock(&ci->i_ceph_lock); |
| 642 | ci->i_inline_version = CEPH_INLINE_NONE; |
| 643 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 644 | &aio_req->prealloc_cf); |
| 645 | spin_unlock(&ci->i_ceph_lock); |
| 646 | if (dirty) |
| 647 | __mark_inode_dirty(inode, dirty); |
| 648 | |
| 649 | } |
| 650 | |
| 651 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : |
| 652 | CEPH_CAP_FILE_RD)); |
| 653 | |
| 654 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); |
| 655 | |
| 656 | ceph_free_cap_flush(aio_req->prealloc_cf); |
| 657 | kfree(aio_req); |
| 658 | } |
| 659 | |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 660 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 661 | { |
| 662 | int rc = req->r_result; |
| 663 | struct inode *inode = req->r_inode; |
| 664 | struct ceph_aio_request *aio_req = req->r_priv; |
| 665 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); |
| 666 | int num_pages = calc_pages_for((u64)osd_data->alignment, |
| 667 | osd_data->length); |
| 668 | |
| 669 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", |
| 670 | inode, rc, osd_data->length); |
| 671 | |
| 672 | if (rc == -EOLDSNAPC) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 673 | struct ceph_aio_work *aio_work; |
| 674 | BUG_ON(!aio_req->write); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 675 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 676 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); |
| 677 | if (aio_work) { |
| 678 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); |
| 679 | aio_work->req = req; |
| 680 | queue_work(ceph_inode_to_client(inode)->wb_wq, |
| 681 | &aio_work->work); |
| 682 | return; |
| 683 | } |
| 684 | rc = -ENOMEM; |
| 685 | } else if (!aio_req->write) { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 686 | if (rc == -ENOENT) |
| 687 | rc = 0; |
| 688 | if (rc >= 0 && osd_data->length > rc) { |
| 689 | int zoff = osd_data->alignment + rc; |
| 690 | int zlen = osd_data->length - rc; |
| 691 | /* |
| 692 | * If read is satisfied by single OSD request, |
| 693 | * it can pass EOF. Otherwise read is within |
| 694 | * i_size. |
| 695 | */ |
| 696 | if (aio_req->num_reqs == 1) { |
| 697 | loff_t i_size = i_size_read(inode); |
| 698 | loff_t endoff = aio_req->iocb->ki_pos + rc; |
| 699 | if (endoff < i_size) |
| 700 | zlen = min_t(size_t, zlen, |
| 701 | i_size - endoff); |
| 702 | aio_req->total_len = rc + zlen; |
| 703 | } |
| 704 | |
| 705 | if (zlen > 0) |
| 706 | ceph_zero_page_vector_range(zoff, zlen, |
| 707 | osd_data->pages); |
| 708 | } |
| 709 | } |
| 710 | |
Yan, Zheng | a22bd5f | 2016-05-26 10:30:13 +0800 | [diff] [blame] | 711 | ceph_put_page_vector(osd_data->pages, num_pages, !aio_req->write); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 712 | ceph_osdc_put_request(req); |
| 713 | |
| 714 | if (rc < 0) |
| 715 | cmpxchg(&aio_req->error, 0, rc); |
| 716 | |
| 717 | ceph_aio_complete(inode, aio_req); |
| 718 | return; |
| 719 | } |
| 720 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 721 | static void ceph_aio_retry_work(struct work_struct *work) |
| 722 | { |
| 723 | struct ceph_aio_work *aio_work = |
| 724 | container_of(work, struct ceph_aio_work, work); |
| 725 | struct ceph_osd_request *orig_req = aio_work->req; |
| 726 | struct ceph_aio_request *aio_req = orig_req->r_priv; |
| 727 | struct inode *inode = orig_req->r_inode; |
| 728 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 729 | struct ceph_snap_context *snapc; |
| 730 | struct ceph_osd_request *req; |
| 731 | int ret; |
| 732 | |
| 733 | spin_lock(&ci->i_ceph_lock); |
| 734 | if (__ceph_have_pending_cap_snap(ci)) { |
| 735 | struct ceph_cap_snap *capsnap = |
| 736 | list_last_entry(&ci->i_cap_snaps, |
| 737 | struct ceph_cap_snap, |
| 738 | ci_item); |
| 739 | snapc = ceph_get_snap_context(capsnap->context); |
| 740 | } else { |
| 741 | BUG_ON(!ci->i_head_snapc); |
| 742 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
| 743 | } |
| 744 | spin_unlock(&ci->i_ceph_lock); |
| 745 | |
| 746 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, |
| 747 | false, GFP_NOFS); |
Dan Carpenter | 1418bf0 | 2016-01-26 12:24:44 +0300 | [diff] [blame] | 748 | if (!req) { |
| 749 | ret = -ENOMEM; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 750 | req = orig_req; |
| 751 | goto out; |
| 752 | } |
| 753 | |
| 754 | req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 755 | CEPH_OSD_FLAG_ONDISK | |
| 756 | CEPH_OSD_FLAG_WRITE; |
Ilya Dryomov | 63244fa | 2016-04-28 16:07:23 +0200 | [diff] [blame] | 757 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
Ilya Dryomov | d30291b | 2016-04-29 19:54:20 +0200 | [diff] [blame] | 758 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 759 | |
Ilya Dryomov | 13d1ad1 | 2016-04-27 14:15:51 +0200 | [diff] [blame] | 760 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
| 761 | if (ret) { |
| 762 | ceph_osdc_put_request(req); |
| 763 | req = orig_req; |
| 764 | goto out; |
| 765 | } |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 766 | |
| 767 | req->r_ops[0] = orig_req->r_ops[0]; |
| 768 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
| 769 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 770 | req->r_mtime = aio_req->mtime; |
| 771 | req->r_data_offset = req->r_ops[0].extent.offset; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 772 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 773 | ceph_osdc_put_request(orig_req); |
| 774 | |
| 775 | req->r_callback = ceph_aio_complete_req; |
| 776 | req->r_inode = inode; |
| 777 | req->r_priv = aio_req; |
| 778 | |
| 779 | ret = ceph_osdc_start_request(req->r_osdc, req, false); |
| 780 | out: |
| 781 | if (ret < 0) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 782 | req->r_result = ret; |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 783 | ceph_aio_complete_req(req); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 784 | } |
| 785 | |
Yan, Zheng | db6aed7 | 2016-01-26 23:05:37 +0800 | [diff] [blame] | 786 | ceph_put_snap_context(snapc); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 787 | kfree(aio_work); |
| 788 | } |
| 789 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 790 | /* |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 791 | * Write commit request unsafe callback, called to tell us when a |
| 792 | * request is unsafe (that is, in flight--has been handed to the |
| 793 | * messenger to send to its target osd). It is called again when |
| 794 | * we've received a response message indicating the request is |
| 795 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request |
| 796 | * is completed early (and unsuccessfully) due to a timeout or |
| 797 | * interrupt. |
| 798 | * |
| 799 | * This is used if we requested both an ACK and ONDISK commit reply |
| 800 | * from the OSD. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 801 | */ |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 802 | static void ceph_sync_write_unsafe(struct ceph_osd_request *req, bool unsafe) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 803 | { |
| 804 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); |
| 805 | |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 806 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
| 807 | unsafe ? "un" : ""); |
| 808 | if (unsafe) { |
| 809 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); |
| 810 | spin_lock(&ci->i_unsafe_lock); |
| 811 | list_add_tail(&req->r_unsafe_item, |
| 812 | &ci->i_unsafe_writes); |
| 813 | spin_unlock(&ci->i_unsafe_lock); |
Ilya Dryomov | fe5da05 | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 814 | |
| 815 | complete_all(&req->r_completion); |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 816 | } else { |
| 817 | spin_lock(&ci->i_unsafe_lock); |
| 818 | list_del_init(&req->r_unsafe_item); |
| 819 | spin_unlock(&ci->i_unsafe_lock); |
| 820 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); |
| 821 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 822 | } |
| 823 | |
Yan, Zheng | 9a5530c | 2016-06-15 16:29:18 +0800 | [diff] [blame^] | 824 | /* |
| 825 | * Wait on any unsafe replies for the given inode. First wait on the |
| 826 | * newest request, and make that the upper bound. Then, if there are |
| 827 | * more requests, keep waiting on the oldest as long as it is still older |
| 828 | * than the original request. |
| 829 | */ |
| 830 | void ceph_sync_write_wait(struct inode *inode) |
| 831 | { |
| 832 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 833 | struct list_head *head = &ci->i_unsafe_writes; |
| 834 | struct ceph_osd_request *req; |
| 835 | u64 last_tid; |
| 836 | |
| 837 | if (!S_ISREG(inode->i_mode)) |
| 838 | return; |
| 839 | |
| 840 | spin_lock(&ci->i_unsafe_lock); |
| 841 | if (list_empty(head)) |
| 842 | goto out; |
| 843 | |
| 844 | /* set upper bound as _last_ entry in chain */ |
| 845 | |
| 846 | req = list_last_entry(head, struct ceph_osd_request, |
| 847 | r_unsafe_item); |
| 848 | last_tid = req->r_tid; |
| 849 | |
| 850 | do { |
| 851 | ceph_osdc_get_request(req); |
| 852 | spin_unlock(&ci->i_unsafe_lock); |
| 853 | |
| 854 | dout("sync_write_wait on tid %llu (until %llu)\n", |
| 855 | req->r_tid, last_tid); |
| 856 | wait_for_completion(&req->r_safe_completion); |
| 857 | ceph_osdc_put_request(req); |
| 858 | |
| 859 | spin_lock(&ci->i_unsafe_lock); |
| 860 | /* |
| 861 | * from here on look at first entry in chain, since we |
| 862 | * only want to wait for anything older than last_tid |
| 863 | */ |
| 864 | if (list_empty(head)) |
| 865 | break; |
| 866 | req = list_first_entry(head, struct ceph_osd_request, |
| 867 | r_unsafe_item); |
| 868 | } while (req->r_tid < last_tid); |
| 869 | out: |
| 870 | spin_unlock(&ci->i_unsafe_lock); |
| 871 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 872 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 873 | static ssize_t |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 874 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
| 875 | struct ceph_snap_context *snapc, |
| 876 | struct ceph_cap_flush **pcf) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 877 | { |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 878 | struct file *file = iocb->ki_filp; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 879 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 880 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 881 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
Alex Elder | acead00 | 2013-03-14 14:09:05 -0500 | [diff] [blame] | 882 | struct ceph_vino vino; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 883 | struct ceph_osd_request *req; |
| 884 | struct page **pages; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 885 | struct ceph_aio_request *aio_req = NULL; |
| 886 | int num_pages = 0; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 887 | int flags; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 888 | int ret; |
Deepa Dinamani | 8bbd471 | 2016-02-02 22:07:48 -0800 | [diff] [blame] | 889 | struct timespec mtime = current_fs_time(inode->i_sb); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 890 | size_t count = iov_iter_count(iter); |
| 891 | loff_t pos = iocb->ki_pos; |
| 892 | bool write = iov_iter_rw(iter) == WRITE; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 893 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 894 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 895 | return -EROFS; |
| 896 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 897 | dout("sync_direct_read_write (%s) on file %p %lld~%u\n", |
| 898 | (write ? "write" : "read"), file, pos, (unsigned)count); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 899 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 900 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 901 | if (ret < 0) |
| 902 | return ret; |
| 903 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 904 | if (write) { |
| 905 | ret = invalidate_inode_pages2_range(inode->i_mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 906 | pos >> PAGE_SHIFT, |
| 907 | (pos + count) >> PAGE_SHIFT); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 908 | if (ret < 0) |
| 909 | dout("invalidate_inode_pages2_range returned %d\n", ret); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 910 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 911 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 912 | CEPH_OSD_FLAG_ONDISK | |
| 913 | CEPH_OSD_FLAG_WRITE; |
| 914 | } else { |
| 915 | flags = CEPH_OSD_FLAG_READ; |
| 916 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 917 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 918 | while (iov_iter_count(iter) > 0) { |
| 919 | u64 size = dio_get_pagev_size(iter); |
| 920 | size_t start = 0; |
| 921 | ssize_t len; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 922 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 923 | vino = ceph_vino(inode); |
| 924 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 925 | vino, pos, &size, 0, |
| 926 | /*include a 'startsync' command*/ |
| 927 | write ? 2 : 1, |
| 928 | write ? CEPH_OSD_OP_WRITE : |
| 929 | CEPH_OSD_OP_READ, |
| 930 | flags, snapc, |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 931 | ci->i_truncate_seq, |
| 932 | ci->i_truncate_size, |
| 933 | false); |
| 934 | if (IS_ERR(req)) { |
| 935 | ret = PTR_ERR(req); |
Al Viro | eab8723 | 2014-04-03 22:44:19 -0400 | [diff] [blame] | 936 | break; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 937 | } |
| 938 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 939 | len = size; |
| 940 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); |
Zhu, Caifeng | b5b98989d | 2015-10-08 15:26:15 +0800 | [diff] [blame] | 941 | if (IS_ERR(pages)) { |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 942 | ceph_osdc_put_request(req); |
Zhu, Caifeng | b5b98989d | 2015-10-08 15:26:15 +0800 | [diff] [blame] | 943 | ret = PTR_ERR(pages); |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 944 | break; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 945 | } |
| 946 | |
| 947 | /* |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 948 | * To simplify error handling, allow AIO when IO within i_size |
| 949 | * or IO can be satisfied by single OSD request. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 950 | */ |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 951 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
| 952 | (len == count || pos + count <= i_size_read(inode))) { |
| 953 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); |
| 954 | if (aio_req) { |
| 955 | aio_req->iocb = iocb; |
| 956 | aio_req->write = write; |
| 957 | INIT_LIST_HEAD(&aio_req->osd_reqs); |
| 958 | if (write) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 959 | aio_req->mtime = mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 960 | swap(aio_req->prealloc_cf, *pcf); |
| 961 | } |
| 962 | } |
| 963 | /* ignore error */ |
| 964 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 965 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 966 | if (write) { |
| 967 | /* |
| 968 | * throw out any page cache pages in this range. this |
| 969 | * may block. |
| 970 | */ |
| 971 | truncate_inode_pages_range(inode->i_mapping, pos, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 972 | (pos+len) | (PAGE_SIZE - 1)); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 973 | |
| 974 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 975 | req->r_mtime = mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 976 | } |
| 977 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 978 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, |
| 979 | false, false); |
| 980 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 981 | if (aio_req) { |
| 982 | aio_req->total_len += len; |
| 983 | aio_req->num_reqs++; |
| 984 | atomic_inc(&aio_req->pending_reqs); |
| 985 | |
| 986 | req->r_callback = ceph_aio_complete_req; |
| 987 | req->r_inode = inode; |
| 988 | req->r_priv = aio_req; |
| 989 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); |
| 990 | |
| 991 | pos += len; |
| 992 | iov_iter_advance(iter, len); |
| 993 | continue; |
| 994 | } |
| 995 | |
| 996 | ret = ceph_osdc_start_request(req->r_osdc, req, false); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 997 | if (!ret) |
| 998 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
| 999 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1000 | size = i_size_read(inode); |
| 1001 | if (!write) { |
| 1002 | if (ret == -ENOENT) |
| 1003 | ret = 0; |
| 1004 | if (ret >= 0 && ret < len && pos + ret < size) { |
| 1005 | int zlen = min_t(size_t, len - ret, |
| 1006 | size - pos - ret); |
| 1007 | ceph_zero_page_vector_range(start + ret, zlen, |
| 1008 | pages); |
| 1009 | ret += zlen; |
| 1010 | } |
| 1011 | if (ret >= 0) |
| 1012 | len = ret; |
| 1013 | } |
| 1014 | |
Yan, Zheng | a22bd5f | 2016-05-26 10:30:13 +0800 | [diff] [blame] | 1015 | ceph_put_page_vector(pages, num_pages, !write); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1016 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1017 | ceph_osdc_put_request(req); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1018 | if (ret < 0) |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1019 | break; |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 1020 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1021 | pos += len; |
| 1022 | iov_iter_advance(iter, len); |
| 1023 | |
| 1024 | if (!write && pos >= size) |
| 1025 | break; |
| 1026 | |
| 1027 | if (write && pos > size) { |
| 1028 | if (ceph_inode_set_size(inode, pos)) |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 1029 | ceph_check_caps(ceph_inode(inode), |
| 1030 | CHECK_CAPS_AUTHONLY, |
| 1031 | NULL); |
| 1032 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1033 | } |
| 1034 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1035 | if (aio_req) { |
Yan, Zheng | fc8c389 | 2016-06-14 11:13:59 +0800 | [diff] [blame] | 1036 | LIST_HEAD(osd_reqs); |
| 1037 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1038 | if (aio_req->num_reqs == 0) { |
| 1039 | kfree(aio_req); |
| 1040 | return ret; |
| 1041 | } |
| 1042 | |
| 1043 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : |
| 1044 | CEPH_CAP_FILE_RD); |
| 1045 | |
Yan, Zheng | fc8c389 | 2016-06-14 11:13:59 +0800 | [diff] [blame] | 1046 | list_splice(&aio_req->osd_reqs, &osd_reqs); |
| 1047 | while (!list_empty(&osd_reqs)) { |
| 1048 | req = list_first_entry(&osd_reqs, |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1049 | struct ceph_osd_request, |
| 1050 | r_unsafe_item); |
| 1051 | list_del_init(&req->r_unsafe_item); |
| 1052 | if (ret >= 0) |
| 1053 | ret = ceph_osdc_start_request(req->r_osdc, |
| 1054 | req, false); |
| 1055 | if (ret < 0) { |
| 1056 | req->r_result = ret; |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 1057 | ceph_aio_complete_req(req); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1058 | } |
| 1059 | } |
| 1060 | return -EIOCBQUEUED; |
| 1061 | } |
| 1062 | |
| 1063 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { |
| 1064 | ret = pos - iocb->ki_pos; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1065 | iocb->ki_pos = pos; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1066 | } |
| 1067 | return ret; |
| 1068 | } |
| 1069 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1070 | /* |
| 1071 | * Synchronous write, straight from __user pointer or user pages. |
| 1072 | * |
| 1073 | * If write spans object boundary, just do multiple writes. (For a |
| 1074 | * correct atomic write, we should e.g. take write locks on all |
| 1075 | * objects, rollback on failure, etc.) |
| 1076 | */ |
Yan, Zheng | 06fee30 | 2014-07-28 14:33:46 +0800 | [diff] [blame] | 1077 | static ssize_t |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1078 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
| 1079 | struct ceph_snap_context *snapc) |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1080 | { |
| 1081 | struct file *file = iocb->ki_filp; |
| 1082 | struct inode *inode = file_inode(file); |
| 1083 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1084 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1085 | struct ceph_vino vino; |
| 1086 | struct ceph_osd_request *req; |
| 1087 | struct page **pages; |
| 1088 | u64 len; |
| 1089 | int num_pages; |
| 1090 | int written = 0; |
| 1091 | int flags; |
| 1092 | int check_caps = 0; |
| 1093 | int ret; |
Deepa Dinamani | 8bbd471 | 2016-02-02 22:07:48 -0800 | [diff] [blame] | 1094 | struct timespec mtime = current_fs_time(inode->i_sb); |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1095 | size_t count = iov_iter_count(from); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1096 | |
| 1097 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
| 1098 | return -EROFS; |
| 1099 | |
| 1100 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); |
| 1101 | |
| 1102 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
| 1103 | if (ret < 0) |
| 1104 | return ret; |
| 1105 | |
| 1106 | ret = invalidate_inode_pages2_range(inode->i_mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1107 | pos >> PAGE_SHIFT, |
| 1108 | (pos + count) >> PAGE_SHIFT); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1109 | if (ret < 0) |
| 1110 | dout("invalidate_inode_pages2_range returned %d\n", ret); |
| 1111 | |
| 1112 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 1113 | CEPH_OSD_FLAG_ONDISK | |
| 1114 | CEPH_OSD_FLAG_WRITE | |
| 1115 | CEPH_OSD_FLAG_ACK; |
| 1116 | |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1117 | while ((len = iov_iter_count(from)) > 0) { |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1118 | size_t left; |
| 1119 | int n; |
| 1120 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1121 | vino = ceph_vino(inode); |
| 1122 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
Yan, Zheng | 715e4cd | 2014-11-13 14:40:37 +0800 | [diff] [blame] | 1123 | vino, pos, &len, 0, 1, |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1124 | CEPH_OSD_OP_WRITE, flags, snapc, |
| 1125 | ci->i_truncate_seq, |
| 1126 | ci->i_truncate_size, |
| 1127 | false); |
| 1128 | if (IS_ERR(req)) { |
| 1129 | ret = PTR_ERR(req); |
Al Viro | eab8723 | 2014-04-03 22:44:19 -0400 | [diff] [blame] | 1130 | break; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1131 | } |
| 1132 | |
| 1133 | /* |
| 1134 | * write from beginning of first page, |
| 1135 | * regardless of io alignment |
| 1136 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1137 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1138 | |
Yan, Zheng | 687265e | 2015-06-13 17:27:05 +0800 | [diff] [blame] | 1139 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1140 | if (IS_ERR(pages)) { |
| 1141 | ret = PTR_ERR(pages); |
| 1142 | goto out; |
| 1143 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1144 | |
| 1145 | left = len; |
| 1146 | for (n = 0; n < num_pages; n++) { |
Ilya Dryomov | 125d725 | 2014-01-28 19:16:18 +0200 | [diff] [blame] | 1147 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1148 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1149 | if (ret != plen) { |
| 1150 | ret = -EFAULT; |
| 1151 | break; |
| 1152 | } |
| 1153 | left -= ret; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1154 | } |
| 1155 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1156 | if (ret < 0) { |
| 1157 | ceph_release_page_vector(pages, num_pages); |
| 1158 | goto out; |
| 1159 | } |
| 1160 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1161 | /* get a second commit callback */ |
| 1162 | req->r_unsafe_callback = ceph_sync_write_unsafe; |
| 1163 | req->r_inode = inode; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1164 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1165 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
| 1166 | false, true); |
Alex Elder | 02ee07d | 2013-03-14 14:09:06 -0500 | [diff] [blame] | 1167 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 1168 | req->r_mtime = mtime; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1169 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
| 1170 | if (!ret) |
| 1171 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1172 | |
| 1173 | out: |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1174 | ceph_osdc_put_request(req); |
| 1175 | if (ret == 0) { |
| 1176 | pos += len; |
| 1177 | written += len; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1178 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1179 | if (pos > i_size_read(inode)) { |
| 1180 | check_caps = ceph_inode_set_size(inode, pos); |
| 1181 | if (check_caps) |
| 1182 | ceph_check_caps(ceph_inode(inode), |
| 1183 | CHECK_CAPS_AUTHONLY, |
| 1184 | NULL); |
| 1185 | } |
| 1186 | } else |
| 1187 | break; |
| 1188 | } |
| 1189 | |
| 1190 | if (ret != -EOLDSNAPC && written > 0) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1191 | ret = written; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1192 | iocb->ki_pos = pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1193 | } |
| 1194 | return ret; |
| 1195 | } |
| 1196 | |
| 1197 | /* |
| 1198 | * Wrap generic_file_aio_read with checks for cap bits on the inode. |
| 1199 | * Atomically grab references, so that those bits are not released |
| 1200 | * back to the MDS mid-read. |
| 1201 | * |
| 1202 | * Hmm, the sync read case isn't actually async... should it be? |
| 1203 | */ |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1204 | static ssize_t ceph_read_iter(struct kiocb *iocb, struct iov_iter *to) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1205 | { |
| 1206 | struct file *filp = iocb->ki_filp; |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1207 | struct ceph_file_info *fi = filp->private_data; |
Christoph Hellwig | 66ee59a | 2015-02-11 19:56:46 +0100 | [diff] [blame] | 1208 | size_t len = iov_iter_count(to); |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1209 | struct inode *inode = file_inode(filp); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1210 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1211 | struct page *pinned_page = NULL; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1212 | ssize_t ret; |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1213 | int want, got = 0; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1214 | int retry_op = 0, read = 0; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1215 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1216 | again: |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1217 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
| 1218 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); |
| 1219 | |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1220 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1221 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; |
| 1222 | else |
| 1223 | want = CEPH_CAP_FILE_CACHE; |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1224 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1225 | if (ret < 0) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1226 | return ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1227 | |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1228 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1229 | (iocb->ki_flags & IOCB_DIRECT) || |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1230 | (fi->flags & CEPH_F_SYNC)) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1231 | |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1232 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
| 1233 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
| 1234 | ceph_cap_string(got)); |
| 1235 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1236 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1237 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
| 1238 | ret = ceph_direct_read_write(iocb, to, |
| 1239 | NULL, NULL); |
| 1240 | if (ret >= 0 && ret < len) |
| 1241 | retry_op = CHECK_EOF; |
| 1242 | } else { |
| 1243 | ret = ceph_sync_read(iocb, to, &retry_op); |
| 1244 | } |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1245 | } else { |
| 1246 | retry_op = READ_INLINE; |
| 1247 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1248 | } else { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1249 | dout("aio_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1250 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1251 | ceph_cap_string(got)); |
| 1252 | |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1253 | ret = generic_file_read_iter(iocb, to); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1254 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1255 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
| 1256 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1257 | if (pinned_page) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1258 | put_page(pinned_page); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1259 | pinned_page = NULL; |
| 1260 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1261 | ceph_put_cap_refs(ci, got); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1262 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1263 | int statret; |
| 1264 | struct page *page = NULL; |
| 1265 | loff_t i_size; |
| 1266 | if (retry_op == READ_INLINE) { |
Yan, Zheng | 687265e | 2015-06-13 17:27:05 +0800 | [diff] [blame] | 1267 | page = __page_cache_alloc(GFP_KERNEL); |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1268 | if (!page) |
| 1269 | return -ENOMEM; |
| 1270 | } |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1271 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1272 | statret = __ceph_do_getattr(inode, page, |
| 1273 | CEPH_STAT_CAP_INLINE_DATA, !!page); |
| 1274 | if (statret < 0) { |
| 1275 | __free_page(page); |
| 1276 | if (statret == -ENODATA) { |
| 1277 | BUG_ON(retry_op != READ_INLINE); |
| 1278 | goto again; |
| 1279 | } |
| 1280 | return statret; |
| 1281 | } |
| 1282 | |
| 1283 | i_size = i_size_read(inode); |
| 1284 | if (retry_op == READ_INLINE) { |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1285 | BUG_ON(ret > 0 || read > 0); |
| 1286 | if (iocb->ki_pos < i_size && |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1287 | iocb->ki_pos < PAGE_SIZE) { |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1288 | loff_t end = min_t(loff_t, i_size, |
| 1289 | iocb->ki_pos + len); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1290 | end = min_t(loff_t, end, PAGE_SIZE); |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1291 | if (statret < end) |
| 1292 | zero_user_segment(page, statret, end); |
| 1293 | ret = copy_page_to_iter(page, |
| 1294 | iocb->ki_pos & ~PAGE_MASK, |
| 1295 | end - iocb->ki_pos, to); |
| 1296 | iocb->ki_pos += ret; |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1297 | read += ret; |
| 1298 | } |
| 1299 | if (iocb->ki_pos < i_size && read < len) { |
| 1300 | size_t zlen = min_t(size_t, len - read, |
| 1301 | i_size - iocb->ki_pos); |
| 1302 | ret = iov_iter_zero(zlen, to); |
| 1303 | iocb->ki_pos += ret; |
| 1304 | read += ret; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1305 | } |
| 1306 | __free_pages(page, 0); |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1307 | return read; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1308 | } |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1309 | |
| 1310 | /* hit EOF or hole? */ |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1311 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1312 | ret < len) { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1313 | dout("sync_read hit hole, ppos %lld < size %lld" |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1314 | ", reading more\n", iocb->ki_pos, i_size); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1315 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1316 | read += ret; |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1317 | len -= ret; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1318 | retry_op = HAVE_RETRIED; |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1319 | goto again; |
| 1320 | } |
| 1321 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1322 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1323 | if (ret >= 0) |
| 1324 | ret += read; |
| 1325 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1326 | return ret; |
| 1327 | } |
| 1328 | |
| 1329 | /* |
| 1330 | * Take cap references to avoid releasing caps to MDS mid-write. |
| 1331 | * |
| 1332 | * If we are synchronous, and write with an old snap context, the OSD |
| 1333 | * may return EOLDSNAPC. In that case, retry the write.. _after_ |
| 1334 | * dropping our cap refs and allowing the pending snap to logically |
| 1335 | * complete _before_ this write occurs. |
| 1336 | * |
| 1337 | * If we are near ENOSPC, write synchronously. |
| 1338 | */ |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1339 | static ssize_t ceph_write_iter(struct kiocb *iocb, struct iov_iter *from) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1340 | { |
| 1341 | struct file *file = iocb->ki_filp; |
Sage Weil | 33caad3 | 2010-05-26 14:31:27 -0700 | [diff] [blame] | 1342 | struct ceph_file_info *fi = file->private_data; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1343 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1344 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 1345 | struct ceph_osd_client *osdc = |
| 1346 | &ceph_sb_to_client(inode->i_sb)->client->osdc; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1347 | struct ceph_cap_flush *prealloc_cf; |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1348 | ssize_t count, written = 0; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1349 | int err, want, got; |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1350 | loff_t pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1351 | |
| 1352 | if (ceph_snap(inode) != CEPH_NOSNAP) |
| 1353 | return -EROFS; |
| 1354 | |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1355 | prealloc_cf = ceph_alloc_cap_flush(); |
| 1356 | if (!prealloc_cf) |
| 1357 | return -ENOMEM; |
| 1358 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1359 | inode_lock(inode); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1360 | |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1361 | /* We can write back this queue in page reclaim */ |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 1362 | current->backing_dev_info = inode_to_bdi(inode); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1363 | |
Yan, Zheng | 55b0b31 | 2015-09-07 11:35:01 +0800 | [diff] [blame] | 1364 | if (iocb->ki_flags & IOCB_APPEND) { |
| 1365 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
| 1366 | if (err < 0) |
| 1367 | goto out; |
| 1368 | } |
| 1369 | |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1370 | err = generic_write_checks(iocb, from); |
| 1371 | if (err <= 0) |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1372 | goto out; |
| 1373 | |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1374 | pos = iocb->ki_pos; |
| 1375 | count = iov_iter_count(from); |
Jan Kara | 5fa8e0a | 2015-05-21 16:05:53 +0200 | [diff] [blame] | 1376 | err = file_remove_privs(file); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1377 | if (err) |
| 1378 | goto out; |
| 1379 | |
| 1380 | err = file_update_time(file); |
| 1381 | if (err) |
| 1382 | goto out; |
| 1383 | |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1384 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
| 1385 | err = ceph_uninline_data(file, NULL); |
| 1386 | if (err < 0) |
| 1387 | goto out; |
| 1388 | } |
| 1389 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1390 | retry_snap: |
Ilya Dryomov | b7ec35b | 2016-04-28 16:07:25 +0200 | [diff] [blame] | 1391 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL)) { |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1392 | err = -ENOSPC; |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1393 | goto out; |
| 1394 | } |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1395 | |
Randy Dunlap | ac7f29b | 2013-04-19 14:20:07 -0700 | [diff] [blame] | 1396 | dout("aio_write %p %llx.%llx %llu~%zd getting caps. i_size %llu\n", |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1397 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1398 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1399 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; |
| 1400 | else |
| 1401 | want = CEPH_CAP_FILE_BUFFER; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1402 | got = 0; |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1403 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
| 1404 | &got, NULL); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1405 | if (err < 0) |
Yan, Zheng | 37505d5 | 2013-04-12 16:11:10 +0800 | [diff] [blame] | 1406 | goto out; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1407 | |
Randy Dunlap | ac7f29b | 2013-04-19 14:20:07 -0700 | [diff] [blame] | 1408 | dout("aio_write %p %llx.%llx %llu~%zd got cap refs on %s\n", |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1409 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1410 | |
| 1411 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1412 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1413 | struct ceph_snap_context *snapc; |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1414 | struct iov_iter data; |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1415 | inode_unlock(inode); |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1416 | |
| 1417 | spin_lock(&ci->i_ceph_lock); |
| 1418 | if (__ceph_have_pending_cap_snap(ci)) { |
| 1419 | struct ceph_cap_snap *capsnap = |
| 1420 | list_last_entry(&ci->i_cap_snaps, |
| 1421 | struct ceph_cap_snap, |
| 1422 | ci_item); |
| 1423 | snapc = ceph_get_snap_context(capsnap->context); |
| 1424 | } else { |
| 1425 | BUG_ON(!ci->i_head_snapc); |
| 1426 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
| 1427 | } |
| 1428 | spin_unlock(&ci->i_ceph_lock); |
| 1429 | |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1430 | /* we might need to revert back to that point */ |
| 1431 | data = *from; |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1432 | if (iocb->ki_flags & IOCB_DIRECT) |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1433 | written = ceph_direct_read_write(iocb, &data, snapc, |
| 1434 | &prealloc_cf); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1435 | else |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1436 | written = ceph_sync_write(iocb, &data, pos, snapc); |
majianpeng | 0e5dd45 | 2013-08-08 15:32:19 +0800 | [diff] [blame] | 1437 | if (written == -EOLDSNAPC) { |
| 1438 | dout("aio_write %p %llx.%llx %llu~%u" |
| 1439 | "got EOLDSNAPC, retrying\n", |
| 1440 | inode, ceph_vinop(inode), |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1441 | pos, (unsigned)count); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1442 | inode_lock(inode); |
majianpeng | 0e5dd45 | 2013-08-08 15:32:19 +0800 | [diff] [blame] | 1443 | goto retry_snap; |
| 1444 | } |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1445 | if (written > 0) |
| 1446 | iov_iter_advance(from, written); |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1447 | ceph_put_snap_context(snapc); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1448 | } else { |
Yan, Zheng | b0d7c22 | 2013-08-12 21:42:15 -0700 | [diff] [blame] | 1449 | /* |
| 1450 | * No need to acquire the i_truncate_mutex. Because |
| 1451 | * the MDS revokes Fwb caps before sending truncate |
| 1452 | * message to us. We can't get Fwb cap while there |
| 1453 | * are pending vmtruncate. So write and vmtruncate |
| 1454 | * can not run at the same time |
| 1455 | */ |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1456 | written = generic_perform_write(file, from, pos); |
Al Viro | aec605f4 | 2014-02-11 22:28:43 -0500 | [diff] [blame] | 1457 | if (likely(written >= 0)) |
| 1458 | iocb->ki_pos = pos + written; |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1459 | inode_unlock(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1460 | } |
Sage Weil | d8de9ab | 2011-07-26 11:27:34 -0700 | [diff] [blame] | 1461 | |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1462 | if (written >= 0) { |
Sage Weil | fca65b4 | 2011-05-04 11:33:47 -0700 | [diff] [blame] | 1463 | int dirty; |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 1464 | spin_lock(&ci->i_ceph_lock); |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1465 | ci->i_inline_version = CEPH_INLINE_NONE; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1466 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 1467 | &prealloc_cf); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 1468 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | fca65b4 | 2011-05-04 11:33:47 -0700 | [diff] [blame] | 1469 | if (dirty) |
| 1470 | __mark_inode_dirty(inode, dirty); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1471 | } |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1472 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1473 | dout("aio_write %p %llx.%llx %llu~%u dropping cap refs on %s\n", |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1474 | inode, ceph_vinop(inode), pos, (unsigned)count, |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1475 | ceph_cap_string(got)); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1476 | ceph_put_cap_refs(ci, got); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1477 | |
Christoph Hellwig | 6aa657c | 2016-04-07 08:52:02 -0700 | [diff] [blame] | 1478 | if (written >= 0) { |
Ilya Dryomov | b7ec35b | 2016-04-28 16:07:25 +0200 | [diff] [blame] | 1479 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_NEARFULL)) |
Christoph Hellwig | 6aa657c | 2016-04-07 08:52:02 -0700 | [diff] [blame] | 1480 | iocb->ki_flags |= IOCB_DSYNC; |
| 1481 | |
| 1482 | written = generic_write_sync(iocb, written); |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1483 | } |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1484 | |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1485 | goto out_unlocked; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1486 | |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1487 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1488 | inode_unlock(inode); |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1489 | out_unlocked: |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1490 | ceph_free_cap_flush(prealloc_cf); |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1491 | current->backing_dev_info = NULL; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1492 | return written ? written : err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1493 | } |
| 1494 | |
| 1495 | /* |
| 1496 | * llseek. be sure to verify file size on SEEK_END. |
| 1497 | */ |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1498 | static loff_t ceph_llseek(struct file *file, loff_t offset, int whence) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1499 | { |
| 1500 | struct inode *inode = file->f_mapping->host; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1501 | loff_t i_size; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1502 | int ret; |
| 1503 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1504 | inode_lock(inode); |
Sage Weil | 6a82c47 | 2011-12-13 09:19:26 -0800 | [diff] [blame] | 1505 | |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1506 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
Yan, Zheng | 508b32d | 2014-09-16 21:46:17 +0800 | [diff] [blame] | 1507 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1508 | if (ret < 0) { |
| 1509 | offset = ret; |
| 1510 | goto out; |
| 1511 | } |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1512 | } |
| 1513 | |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1514 | i_size = i_size_read(inode); |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1515 | switch (whence) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1516 | case SEEK_END: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1517 | offset += i_size; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1518 | break; |
| 1519 | case SEEK_CUR: |
| 1520 | /* |
| 1521 | * Here we special-case the lseek(fd, 0, SEEK_CUR) |
| 1522 | * position-querying operation. Avoid rewriting the "same" |
| 1523 | * f_pos value back to the file because a concurrent read(), |
| 1524 | * write() or lseek() might have altered it |
| 1525 | */ |
| 1526 | if (offset == 0) { |
| 1527 | offset = file->f_pos; |
| 1528 | goto out; |
| 1529 | } |
| 1530 | offset += file->f_pos; |
| 1531 | break; |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1532 | case SEEK_DATA: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1533 | if (offset >= i_size) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1534 | ret = -ENXIO; |
| 1535 | goto out; |
| 1536 | } |
| 1537 | break; |
| 1538 | case SEEK_HOLE: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1539 | if (offset >= i_size) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1540 | ret = -ENXIO; |
| 1541 | goto out; |
| 1542 | } |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1543 | offset = i_size; |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1544 | break; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1545 | } |
| 1546 | |
Jie Liu | 46a1c2c | 2013-06-25 12:02:13 +0800 | [diff] [blame] | 1547 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1548 | |
| 1549 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1550 | inode_unlock(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1551 | return offset; |
| 1552 | } |
| 1553 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1554 | static inline void ceph_zero_partial_page( |
| 1555 | struct inode *inode, loff_t offset, unsigned size) |
| 1556 | { |
| 1557 | struct page *page; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1558 | pgoff_t index = offset >> PAGE_SHIFT; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1559 | |
| 1560 | page = find_lock_page(inode->i_mapping, index); |
| 1561 | if (page) { |
| 1562 | wait_on_page_writeback(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1563 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1564 | unlock_page(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1565 | put_page(page); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1566 | } |
| 1567 | } |
| 1568 | |
| 1569 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, |
| 1570 | loff_t length) |
| 1571 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1572 | loff_t nearly = round_up(offset, PAGE_SIZE); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1573 | if (offset < nearly) { |
| 1574 | loff_t size = nearly - offset; |
| 1575 | if (length < size) |
| 1576 | size = length; |
| 1577 | ceph_zero_partial_page(inode, offset, size); |
| 1578 | offset += size; |
| 1579 | length -= size; |
| 1580 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1581 | if (length >= PAGE_SIZE) { |
| 1582 | loff_t size = round_down(length, PAGE_SIZE); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1583 | truncate_pagecache_range(inode, offset, offset + size - 1); |
| 1584 | offset += size; |
| 1585 | length -= size; |
| 1586 | } |
| 1587 | if (length) |
| 1588 | ceph_zero_partial_page(inode, offset, length); |
| 1589 | } |
| 1590 | |
| 1591 | static int ceph_zero_partial_object(struct inode *inode, |
| 1592 | loff_t offset, loff_t *length) |
| 1593 | { |
| 1594 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1595 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
| 1596 | struct ceph_osd_request *req; |
| 1597 | int ret = 0; |
| 1598 | loff_t zero = 0; |
| 1599 | int op; |
| 1600 | |
| 1601 | if (!length) { |
| 1602 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; |
| 1603 | length = &zero; |
| 1604 | } else { |
| 1605 | op = CEPH_OSD_OP_ZERO; |
| 1606 | } |
| 1607 | |
| 1608 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
| 1609 | ceph_vino(inode), |
| 1610 | offset, length, |
Yan, Zheng | 715e4cd | 2014-11-13 14:40:37 +0800 | [diff] [blame] | 1611 | 0, 1, op, |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1612 | CEPH_OSD_FLAG_WRITE | |
| 1613 | CEPH_OSD_FLAG_ONDISK, |
| 1614 | NULL, 0, 0, false); |
| 1615 | if (IS_ERR(req)) { |
| 1616 | ret = PTR_ERR(req); |
| 1617 | goto out; |
| 1618 | } |
| 1619 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 1620 | req->r_mtime = inode->i_mtime; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1621 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
| 1622 | if (!ret) { |
| 1623 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
| 1624 | if (ret == -ENOENT) |
| 1625 | ret = 0; |
| 1626 | } |
| 1627 | ceph_osdc_put_request(req); |
| 1628 | |
| 1629 | out: |
| 1630 | return ret; |
| 1631 | } |
| 1632 | |
| 1633 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) |
| 1634 | { |
| 1635 | int ret = 0; |
| 1636 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yan, Zheng | 7627151 | 2016-02-03 21:24:49 +0800 | [diff] [blame] | 1637 | s32 stripe_unit = ci->i_layout.stripe_unit; |
| 1638 | s32 stripe_count = ci->i_layout.stripe_count; |
| 1639 | s32 object_size = ci->i_layout.object_size; |
Sage Weil | b314a90 | 2013-08-27 12:15:16 -0700 | [diff] [blame] | 1640 | u64 object_set_size = object_size * stripe_count; |
| 1641 | u64 nearly, t; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1642 | |
Sage Weil | b314a90 | 2013-08-27 12:15:16 -0700 | [diff] [blame] | 1643 | /* round offset up to next period boundary */ |
| 1644 | nearly = offset + object_set_size - 1; |
| 1645 | t = nearly; |
| 1646 | nearly -= do_div(t, object_set_size); |
| 1647 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1648 | while (length && offset < nearly) { |
| 1649 | loff_t size = length; |
| 1650 | ret = ceph_zero_partial_object(inode, offset, &size); |
| 1651 | if (ret < 0) |
| 1652 | return ret; |
| 1653 | offset += size; |
| 1654 | length -= size; |
| 1655 | } |
| 1656 | while (length >= object_set_size) { |
| 1657 | int i; |
| 1658 | loff_t pos = offset; |
| 1659 | for (i = 0; i < stripe_count; ++i) { |
| 1660 | ret = ceph_zero_partial_object(inode, pos, NULL); |
| 1661 | if (ret < 0) |
| 1662 | return ret; |
| 1663 | pos += stripe_unit; |
| 1664 | } |
| 1665 | offset += object_set_size; |
| 1666 | length -= object_set_size; |
| 1667 | } |
| 1668 | while (length) { |
| 1669 | loff_t size = length; |
| 1670 | ret = ceph_zero_partial_object(inode, offset, &size); |
| 1671 | if (ret < 0) |
| 1672 | return ret; |
| 1673 | offset += size; |
| 1674 | length -= size; |
| 1675 | } |
| 1676 | return ret; |
| 1677 | } |
| 1678 | |
| 1679 | static long ceph_fallocate(struct file *file, int mode, |
| 1680 | loff_t offset, loff_t length) |
| 1681 | { |
| 1682 | struct ceph_file_info *fi = file->private_data; |
Libo Chen | aa8b60e | 2013-12-11 13:49:11 +0800 | [diff] [blame] | 1683 | struct inode *inode = file_inode(file); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1684 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1685 | struct ceph_osd_client *osdc = |
| 1686 | &ceph_inode_to_client(inode)->client->osdc; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1687 | struct ceph_cap_flush *prealloc_cf; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1688 | int want, got = 0; |
| 1689 | int dirty; |
| 1690 | int ret = 0; |
| 1691 | loff_t endoff = 0; |
| 1692 | loff_t size; |
| 1693 | |
Yan, Zheng | 494d77b | 2014-06-26 15:25:17 +0800 | [diff] [blame] | 1694 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
| 1695 | return -EOPNOTSUPP; |
| 1696 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1697 | if (!S_ISREG(inode->i_mode)) |
| 1698 | return -EOPNOTSUPP; |
| 1699 | |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1700 | prealloc_cf = ceph_alloc_cap_flush(); |
| 1701 | if (!prealloc_cf) |
| 1702 | return -ENOMEM; |
| 1703 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1704 | inode_lock(inode); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1705 | |
| 1706 | if (ceph_snap(inode) != CEPH_NOSNAP) { |
| 1707 | ret = -EROFS; |
| 1708 | goto unlock; |
| 1709 | } |
| 1710 | |
Ilya Dryomov | b7ec35b | 2016-04-28 16:07:25 +0200 | [diff] [blame] | 1711 | if (ceph_osdmap_flag(osdc, CEPH_OSDMAP_FULL) && |
| 1712 | !(mode & FALLOC_FL_PUNCH_HOLE)) { |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1713 | ret = -ENOSPC; |
| 1714 | goto unlock; |
| 1715 | } |
| 1716 | |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1717 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
| 1718 | ret = ceph_uninline_data(file, NULL); |
| 1719 | if (ret < 0) |
| 1720 | goto unlock; |
| 1721 | } |
| 1722 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1723 | size = i_size_read(inode); |
| 1724 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
| 1725 | endoff = offset + length; |
| 1726 | |
| 1727 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1728 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; |
| 1729 | else |
| 1730 | want = CEPH_CAP_FILE_BUFFER; |
| 1731 | |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1732 | ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, endoff, &got, NULL); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1733 | if (ret < 0) |
| 1734 | goto unlock; |
| 1735 | |
| 1736 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
| 1737 | if (offset < size) |
| 1738 | ceph_zero_pagecache_range(inode, offset, length); |
| 1739 | ret = ceph_zero_objects(inode, offset, length); |
| 1740 | } else if (endoff > size) { |
| 1741 | truncate_pagecache_range(inode, size, -1); |
| 1742 | if (ceph_inode_set_size(inode, endoff)) |
| 1743 | ceph_check_caps(ceph_inode(inode), |
| 1744 | CHECK_CAPS_AUTHONLY, NULL); |
| 1745 | } |
| 1746 | |
| 1747 | if (!ret) { |
| 1748 | spin_lock(&ci->i_ceph_lock); |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1749 | ci->i_inline_version = CEPH_INLINE_NONE; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1750 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 1751 | &prealloc_cf); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1752 | spin_unlock(&ci->i_ceph_lock); |
| 1753 | if (dirty) |
| 1754 | __mark_inode_dirty(inode, dirty); |
| 1755 | } |
| 1756 | |
| 1757 | ceph_put_cap_refs(ci, got); |
| 1758 | unlock: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1759 | inode_unlock(inode); |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1760 | ceph_free_cap_flush(prealloc_cf); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1761 | return ret; |
| 1762 | } |
| 1763 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1764 | const struct file_operations ceph_file_fops = { |
| 1765 | .open = ceph_open, |
| 1766 | .release = ceph_release, |
| 1767 | .llseek = ceph_llseek, |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1768 | .read_iter = ceph_read_iter, |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1769 | .write_iter = ceph_write_iter, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1770 | .mmap = ceph_mmap, |
| 1771 | .fsync = ceph_fsync, |
Greg Farnum | 40819f6 | 2010-08-02 15:34:23 -0700 | [diff] [blame] | 1772 | .lock = ceph_lock, |
| 1773 | .flock = ceph_flock, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1774 | .splice_read = generic_file_splice_read, |
Al Viro | 3551dd7 | 2014-04-05 04:40:12 -0400 | [diff] [blame] | 1775 | .splice_write = iter_file_splice_write, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1776 | .unlocked_ioctl = ceph_ioctl, |
| 1777 | .compat_ioctl = ceph_ioctl, |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1778 | .fallocate = ceph_fallocate, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1779 | }; |
| 1780 | |