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 | b5b9898 | 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; |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 140 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 141 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
| 142 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 143 | |
| 144 | switch (inode->i_mode & S_IFMT) { |
| 145 | case S_IFREG: |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 146 | /* First file open request creates the cookie, we want to keep |
| 147 | * this cookie around for the filetime of the inode as not to |
| 148 | * have to worry about fscache register / revoke / operation |
| 149 | * races. |
| 150 | * |
| 151 | * Also, if we know the operation is going to invalidate data |
| 152 | * (non readonly) just nuke the cache right away. |
| 153 | */ |
| 154 | ceph_fscache_register_inode_cookie(mdsc->fsc, ci); |
| 155 | if ((fmode & CEPH_FILE_MODE_WR)) |
| 156 | ceph_fscache_invalidate(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 157 | case S_IFDIR: |
| 158 | dout("init_file %p %p 0%o (regular)\n", inode, file, |
| 159 | inode->i_mode); |
Geliang Tang | 99ec269 | 2016-03-13 15:26:29 +0800 | [diff] [blame] | 160 | cf = kmem_cache_zalloc(ceph_file_cachep, GFP_KERNEL); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 161 | if (cf == NULL) { |
| 162 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 163 | return -ENOMEM; |
| 164 | } |
| 165 | cf->fmode = fmode; |
| 166 | cf->next_offset = 2; |
Yan, Zheng | fdd4e15 | 2015-06-16 20:48:56 +0800 | [diff] [blame] | 167 | cf->readdir_cache_idx = -1; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 168 | file->private_data = cf; |
| 169 | BUG_ON(inode->i_fop->release != ceph_release); |
| 170 | break; |
| 171 | |
| 172 | case S_IFLNK: |
| 173 | dout("init_file %p %p 0%o (symlink)\n", inode, file, |
| 174 | inode->i_mode); |
| 175 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 176 | break; |
| 177 | |
| 178 | default: |
| 179 | dout("init_file %p %p 0%o (special)\n", inode, file, |
| 180 | inode->i_mode); |
| 181 | /* |
| 182 | * we need to drop the open ref now, since we don't |
| 183 | * have .release set to ceph_release. |
| 184 | */ |
| 185 | ceph_put_fmode(ceph_inode(inode), fmode); /* clean up */ |
| 186 | BUG_ON(inode->i_fop->release == ceph_release); |
| 187 | |
| 188 | /* call the proper open fop */ |
| 189 | ret = inode->i_fop->open(inode, file); |
| 190 | } |
| 191 | return ret; |
| 192 | } |
| 193 | |
| 194 | /* |
Yan, Zheng | 7731032 | 2016-04-08 15:27:16 +0800 | [diff] [blame^] | 195 | * try renew caps after session gets killed. |
| 196 | */ |
| 197 | int ceph_renew_caps(struct inode *inode) |
| 198 | { |
| 199 | struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc; |
| 200 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 201 | struct ceph_mds_request *req; |
| 202 | int err, flags, wanted; |
| 203 | |
| 204 | spin_lock(&ci->i_ceph_lock); |
| 205 | wanted = __ceph_caps_file_wanted(ci); |
| 206 | if (__ceph_is_any_real_caps(ci) && |
| 207 | (!(wanted & CEPH_CAP_ANY_WR) == 0 || ci->i_auth_cap)) { |
| 208 | int issued = __ceph_caps_issued(ci, NULL); |
| 209 | spin_unlock(&ci->i_ceph_lock); |
| 210 | dout("renew caps %p want %s issued %s updating mds_wanted\n", |
| 211 | inode, ceph_cap_string(wanted), ceph_cap_string(issued)); |
| 212 | ceph_check_caps(ci, 0, NULL); |
| 213 | return 0; |
| 214 | } |
| 215 | spin_unlock(&ci->i_ceph_lock); |
| 216 | |
| 217 | flags = 0; |
| 218 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) |
| 219 | flags = O_RDWR; |
| 220 | else if (wanted & CEPH_CAP_FILE_RD) |
| 221 | flags = O_RDONLY; |
| 222 | else if (wanted & CEPH_CAP_FILE_WR) |
| 223 | flags = O_WRONLY; |
| 224 | #ifdef O_LAZY |
| 225 | if (wanted & CEPH_CAP_FILE_LAZYIO) |
| 226 | flags |= O_LAZY; |
| 227 | #endif |
| 228 | |
| 229 | req = prepare_open_request(inode->i_sb, flags, 0); |
| 230 | if (IS_ERR(req)) { |
| 231 | err = PTR_ERR(req); |
| 232 | goto out; |
| 233 | } |
| 234 | |
| 235 | req->r_inode = inode; |
| 236 | ihold(inode); |
| 237 | req->r_num_caps = 1; |
| 238 | req->r_fmode = -1; |
| 239 | |
| 240 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
| 241 | ceph_mdsc_put_request(req); |
| 242 | out: |
| 243 | dout("renew caps %p open result=%d\n", inode, err); |
| 244 | return err < 0 ? err : 0; |
| 245 | } |
| 246 | |
| 247 | /* |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 248 | * If we already have the requisite capabilities, we can satisfy |
| 249 | * the open request locally (no need to request new caps from the |
| 250 | * MDS). We do, however, need to inform the MDS (asynchronously) |
| 251 | * if our wanted caps set expands. |
| 252 | */ |
| 253 | int ceph_open(struct inode *inode, struct file *file) |
| 254 | { |
| 255 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 256 | struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb); |
| 257 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 258 | struct ceph_mds_request *req; |
| 259 | struct ceph_file_info *cf = file->private_data; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 260 | int err; |
| 261 | int flags, fmode, wanted; |
| 262 | |
| 263 | if (cf) { |
| 264 | dout("open file %p is already opened\n", file); |
| 265 | return 0; |
| 266 | } |
| 267 | |
| 268 | /* filter out O_CREAT|O_EXCL; vfs did that already. yuck. */ |
| 269 | flags = file->f_flags & ~(O_CREAT|O_EXCL); |
| 270 | if (S_ISDIR(inode->i_mode)) |
| 271 | flags = O_DIRECTORY; /* mds likes to know */ |
| 272 | |
| 273 | dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode, |
| 274 | ceph_vinop(inode), file, flags, file->f_flags); |
| 275 | fmode = ceph_flags_to_mode(flags); |
| 276 | wanted = ceph_caps_for_mode(fmode); |
| 277 | |
| 278 | /* snapped files are read-only */ |
| 279 | if (ceph_snap(inode) != CEPH_NOSNAP && (file->f_mode & FMODE_WRITE)) |
| 280 | return -EROFS; |
| 281 | |
| 282 | /* trivially open snapdir */ |
| 283 | if (ceph_snap(inode) == CEPH_SNAPDIR) { |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 284 | spin_lock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 285 | __ceph_get_fmode(ci, fmode); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 286 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 287 | return ceph_init_file(inode, file, fmode); |
| 288 | } |
| 289 | |
| 290 | /* |
Sage Weil | 7421ab8 | 2010-11-07 09:07:15 -0800 | [diff] [blame] | 291 | * No need to block if we have caps on the auth MDS (for |
| 292 | * write) or any MDS (for read). Update wanted set |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 293 | * asynchronously. |
| 294 | */ |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 295 | spin_lock(&ci->i_ceph_lock); |
Sage Weil | 7421ab8 | 2010-11-07 09:07:15 -0800 | [diff] [blame] | 296 | if (__ceph_is_any_real_caps(ci) && |
| 297 | (((fmode & CEPH_FILE_MODE_WR) == 0) || ci->i_auth_cap)) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 298 | int mds_wanted = __ceph_caps_mds_wanted(ci); |
| 299 | int issued = __ceph_caps_issued(ci, NULL); |
| 300 | |
| 301 | dout("open %p fmode %d want %s issued %s using existing\n", |
| 302 | inode, fmode, ceph_cap_string(wanted), |
| 303 | ceph_cap_string(issued)); |
| 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 | |
| 307 | /* adjust wanted? */ |
| 308 | if ((issued & wanted) != wanted && |
| 309 | (mds_wanted & wanted) != wanted && |
| 310 | ceph_snap(inode) != CEPH_SNAPDIR) |
| 311 | ceph_check_caps(ci, 0, NULL); |
| 312 | |
| 313 | return ceph_init_file(inode, file, fmode); |
| 314 | } else if (ceph_snap(inode) != CEPH_NOSNAP && |
| 315 | (ci->i_snap_caps & wanted) == wanted) { |
| 316 | __ceph_get_fmode(ci, fmode); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 317 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 318 | return ceph_init_file(inode, file, fmode); |
| 319 | } |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 320 | |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 321 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 322 | |
| 323 | dout("open fmode %d wants %s\n", fmode, ceph_cap_string(wanted)); |
| 324 | req = prepare_open_request(inode->i_sb, flags, 0); |
| 325 | if (IS_ERR(req)) { |
| 326 | err = PTR_ERR(req); |
| 327 | goto out; |
| 328 | } |
Sage Weil | 70b666c | 2011-05-27 09:24:26 -0700 | [diff] [blame] | 329 | req->r_inode = inode; |
| 330 | ihold(inode); |
Milosz Tanski | 99ccbd2 | 2013-08-21 17:29:54 -0400 | [diff] [blame] | 331 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 332 | req->r_num_caps = 1; |
Jianpeng Ma | e36d571 | 2015-08-18 10:25:35 +0800 | [diff] [blame] | 333 | err = ceph_mdsc_do_request(mdsc, NULL, req); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 334 | if (!err) |
| 335 | err = ceph_init_file(inode, file, req->r_fmode); |
| 336 | ceph_mdsc_put_request(req); |
| 337 | dout("open result=%d on %llx.%llx\n", err, ceph_vinop(inode)); |
| 338 | out: |
| 339 | return err; |
| 340 | } |
| 341 | |
| 342 | |
| 343 | /* |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 344 | * Do a lookup + open with a single request. If we get a non-existent |
| 345 | * file or symlink, return 1 so the VFS can retry. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 346 | */ |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 347 | int ceph_atomic_open(struct inode *dir, struct dentry *dentry, |
Al Viro | 30d9049 | 2012-06-22 12:40:19 +0400 | [diff] [blame] | 348 | struct file *file, unsigned flags, umode_t mode, |
Al Viro | d958527 | 2012-06-22 12:39:14 +0400 | [diff] [blame] | 349 | int *opened) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 350 | { |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 351 | struct ceph_fs_client *fsc = ceph_sb_to_client(dir->i_sb); |
| 352 | struct ceph_mds_client *mdsc = fsc->mdsc; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 353 | struct ceph_mds_request *req; |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 354 | struct dentry *dn; |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 355 | struct ceph_acls_info acls = {}; |
Yan, Zheng | 315f240 | 2016-03-07 10:34:50 +0800 | [diff] [blame] | 356 | int mask; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 357 | int err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 358 | |
Al Viro | a455589 | 2014-10-21 20:11:25 -0400 | [diff] [blame] | 359 | dout("atomic_open %p dentry %p '%pd' %s flags %d mode 0%o\n", |
| 360 | dir, dentry, dentry, |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 361 | d_unhashed(dentry) ? "unhashed" : "hashed", flags, mode); |
| 362 | |
| 363 | if (dentry->d_name.len > NAME_MAX) |
| 364 | return -ENAMETOOLONG; |
| 365 | |
| 366 | err = ceph_init_dentry(dentry); |
| 367 | if (err < 0) |
| 368 | return err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 369 | |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 370 | if (flags & O_CREAT) { |
| 371 | err = ceph_pre_init_acls(dir, &mode, &acls); |
| 372 | if (err < 0) |
| 373 | return err; |
| 374 | } |
| 375 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 376 | /* do the open */ |
| 377 | req = prepare_open_request(dir->i_sb, flags, mode); |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 378 | if (IS_ERR(req)) { |
| 379 | err = PTR_ERR(req); |
| 380 | goto out_acl; |
| 381 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 382 | req->r_dentry = dget(dentry); |
| 383 | req->r_num_caps = 2; |
| 384 | if (flags & O_CREAT) { |
| 385 | req->r_dentry_drop = CEPH_CAP_FILE_SHARED; |
| 386 | req->r_dentry_unless = CEPH_CAP_FILE_EXCL; |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 387 | if (acls.pagelist) { |
| 388 | req->r_pagelist = acls.pagelist; |
| 389 | acls.pagelist = NULL; |
| 390 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 391 | } |
Yan, Zheng | 315f240 | 2016-03-07 10:34:50 +0800 | [diff] [blame] | 392 | |
| 393 | mask = CEPH_STAT_CAP_INODE | CEPH_CAP_AUTH_SHARED; |
| 394 | if (ceph_security_xattr_wanted(dir)) |
| 395 | mask |= CEPH_CAP_XATTR_SHARED; |
| 396 | req->r_args.open.mask = cpu_to_le32(mask); |
| 397 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 398 | req->r_locked_dir = dir; /* caller holds dir->i_mutex */ |
Sage Weil | acda765 | 2011-07-26 11:27:48 -0700 | [diff] [blame] | 399 | err = ceph_mdsc_do_request(mdsc, |
| 400 | (flags & (O_CREAT|O_TRUNC)) ? dir : NULL, |
| 401 | req); |
Yan, Zheng | bf91c31 | 2015-01-19 13:23:20 +0800 | [diff] [blame] | 402 | err = ceph_handle_snapdir(req, dentry, err); |
Sam Lang | 79aec984 | 2012-12-19 09:44:23 -1000 | [diff] [blame] | 403 | if (err) |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 404 | goto out_req; |
Sam Lang | 79aec984 | 2012-12-19 09:44:23 -1000 | [diff] [blame] | 405 | |
Jianpeng Ma | a43137f | 2015-08-18 10:23:50 +0800 | [diff] [blame] | 406 | if ((flags & O_CREAT) && !req->r_reply_info.head->is_dentry) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 407 | err = ceph_handle_notrace_create(dir, dentry); |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 408 | |
| 409 | if (d_unhashed(dentry)) { |
| 410 | dn = ceph_finish_lookup(req, dentry, err); |
| 411 | if (IS_ERR(dn)) |
| 412 | err = PTR_ERR(dn); |
| 413 | } else { |
| 414 | /* we were given a hashed negative dentry */ |
| 415 | dn = NULL; |
| 416 | } |
Sage Weil | 468640e | 2011-07-26 11:28:11 -0700 | [diff] [blame] | 417 | if (err) |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 418 | goto out_req; |
David Howells | 2b0143b | 2015-03-17 22:25:59 +0000 | [diff] [blame] | 419 | if (dn || d_really_is_negative(dentry) || d_is_symlink(dentry)) { |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 420 | /* make vfs retry on splice, ENOENT, or symlink */ |
| 421 | dout("atomic_open finish_no_open on dn %p\n", dn); |
| 422 | err = finish_no_open(file, dn); |
| 423 | } else { |
| 424 | dout("atomic_open finish_open on dn %p\n", dn); |
Sam Lang | 6e8575f | 2012-12-28 09:56:46 -0800 | [diff] [blame] | 425 | 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] | 426 | ceph_init_inode_acls(d_inode(dentry), &acls); |
Sam Lang | 6e8575f | 2012-12-28 09:56:46 -0800 | [diff] [blame] | 427 | *opened |= FILE_CREATED; |
| 428 | } |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 429 | err = finish_open(file, dentry, ceph_open, opened); |
| 430 | } |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 431 | out_req: |
Yan, Zheng | ab86654 | 2014-04-01 20:20:17 +0800 | [diff] [blame] | 432 | if (!req->r_err && req->r_target_inode) |
| 433 | ceph_put_fmode(ceph_inode(req->r_target_inode), req->r_fmode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 434 | ceph_mdsc_put_request(req); |
Yan, Zheng | b1ee94a | 2014-09-16 20:35:17 +0800 | [diff] [blame] | 435 | out_acl: |
| 436 | ceph_release_acls_info(&acls); |
Sage Weil | 5ef50c3 | 2012-07-31 11:27:36 -0700 | [diff] [blame] | 437 | dout("atomic_open result=%d\n", err); |
Al Viro | d958527 | 2012-06-22 12:39:14 +0400 | [diff] [blame] | 438 | return err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 439 | } |
| 440 | |
| 441 | int ceph_release(struct inode *inode, struct file *file) |
| 442 | { |
| 443 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 444 | struct ceph_file_info *cf = file->private_data; |
| 445 | |
| 446 | dout("release inode %p file %p\n", inode, file); |
| 447 | ceph_put_fmode(ci, cf->fmode); |
| 448 | if (cf->last_readdir) |
| 449 | ceph_mdsc_put_request(cf->last_readdir); |
| 450 | kfree(cf->last_name); |
| 451 | kfree(cf->dir_info); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 452 | kmem_cache_free(ceph_file_cachep, cf); |
Sage Weil | 195d3ce | 2010-03-01 09:57:54 -0800 | [diff] [blame] | 453 | |
| 454 | /* wake up anyone waiting for caps on this inode */ |
Yehuda Sadeh | 03066f2 | 2010-07-27 13:11:08 -0700 | [diff] [blame] | 455 | wake_up_all(&ci->i_cap_wq); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 456 | return 0; |
| 457 | } |
| 458 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 459 | enum { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 460 | HAVE_RETRIED = 1, |
| 461 | CHECK_EOF = 2, |
| 462 | READ_INLINE = 3, |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 463 | }; |
| 464 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 465 | /* |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 466 | * Read a range of bytes striped over one or more objects. Iterate over |
| 467 | * objects we stripe over. (That's not atomic, but good enough for now.) |
| 468 | * |
| 469 | * If we get a short result from the OSD, check against i_size; we need to |
| 470 | * only return a short read to the caller if we hit EOF. |
| 471 | */ |
| 472 | static int striped_read(struct inode *inode, |
| 473 | u64 off, u64 len, |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 474 | struct page **pages, int num_pages, |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 475 | int *checkeof) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 476 | { |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 477 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 478 | struct ceph_inode_info *ci = ceph_inode(inode); |
Dan Carpenter | 688bac4 | 2013-07-23 16:48:01 +0300 | [diff] [blame] | 479 | u64 pos, this_len, left; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 480 | loff_t i_size; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 481 | int page_align, pages_left; |
| 482 | int read, ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 483 | struct page **page_pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 484 | bool hit_stripe, was_short; |
| 485 | |
| 486 | /* |
| 487 | * we may need to do multiple reads. not atomic, unfortunately. |
| 488 | */ |
| 489 | pos = off; |
| 490 | left = len; |
| 491 | page_pos = pages; |
| 492 | pages_left = num_pages; |
| 493 | read = 0; |
| 494 | |
| 495 | more: |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 496 | page_align = pos & ~PAGE_MASK; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 497 | this_len = left; |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 498 | ret = ceph_osdc_readpages(&fsc->client->osdc, ceph_vino(inode), |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 499 | &ci->i_layout, pos, &this_len, |
| 500 | ci->i_truncate_seq, |
| 501 | ci->i_truncate_size, |
Sage Weil | b7495fc | 2010-11-09 12:43:12 -0800 | [diff] [blame] | 502 | page_pos, pages_left, page_align); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 503 | if (ret == -ENOENT) |
| 504 | ret = 0; |
Sage Weil | 0e98728 | 2011-06-07 20:40:35 -0700 | [diff] [blame] | 505 | hit_stripe = this_len < left; |
| 506 | was_short = ret >= 0 && ret < this_len; |
Dan Carpenter | 688bac4 | 2013-07-23 16:48:01 +0300 | [diff] [blame] | 507 | 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] | 508 | ret, hit_stripe ? " HITSTRIPE" : "", was_short ? " SHORT" : ""); |
| 509 | |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 510 | i_size = i_size_read(inode); |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 511 | if (ret >= 0) { |
| 512 | int didpages; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 513 | if (was_short && (pos + ret < i_size)) { |
| 514 | int zlen = min(this_len - ret, i_size - pos - ret); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 515 | int zoff = (off & ~PAGE_MASK) + read + ret; |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 516 | dout(" zero gap %llu to %llu\n", |
Yan, Zheng | 1487a68 | 2015-01-06 15:29:14 +0800 | [diff] [blame] | 517 | pos + ret, pos + ret + zlen); |
| 518 | ceph_zero_page_vector_range(zoff, zlen, pages); |
| 519 | ret += zlen; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 520 | } |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 521 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 522 | didpages = (page_align + ret) >> PAGE_SHIFT; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 523 | pos += ret; |
| 524 | read = pos - off; |
| 525 | left -= ret; |
| 526 | page_pos += didpages; |
| 527 | pages_left -= didpages; |
| 528 | |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 529 | /* hit stripe and need continue*/ |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 530 | if (left && hit_stripe && pos < i_size) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 531 | goto more; |
| 532 | } |
| 533 | |
majianpeng | ee7289b | 2013-08-21 15:02:51 +0800 | [diff] [blame] | 534 | if (read > 0) { |
majianpeng | 02ae66d | 2013-08-06 16:20:38 +0800 | [diff] [blame] | 535 | ret = read; |
Sage Weil | c3cd628 | 2011-06-01 16:08:44 -0700 | [diff] [blame] | 536 | /* did we bounce off eof? */ |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 537 | if (pos + left > i_size) |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 538 | *checkeof = CHECK_EOF; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 539 | } |
| 540 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 541 | dout("striped_read returns %d\n", ret); |
| 542 | return ret; |
| 543 | } |
| 544 | |
| 545 | /* |
| 546 | * Completely synchronous read and write methods. Direct from __user |
| 547 | * buffer to osd, or directly to user pages (if O_DIRECT). |
| 548 | * |
| 549 | * If the read spans object boundary, just do multiple reads. |
| 550 | */ |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 551 | static ssize_t ceph_sync_read(struct kiocb *iocb, struct iov_iter *i, |
| 552 | int *checkeof) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 553 | { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 554 | struct file *file = iocb->ki_filp; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 555 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 556 | struct page **pages; |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 557 | u64 off = iocb->ki_pos; |
Henry C Chang | ab226e2 | 2010-12-15 20:41:54 -0800 | [diff] [blame] | 558 | int num_pages, ret; |
Al Viro | 2b777c9 | 2014-04-03 22:31:22 -0400 | [diff] [blame] | 559 | size_t len = iov_iter_count(i); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 560 | |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 561 | dout("sync_read on file %p %llu~%u %s\n", file, off, |
| 562 | (unsigned)len, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 563 | (file->f_flags & O_DIRECT) ? "O_DIRECT" : ""); |
Yan, Zheng | d0d0db2 | 2014-07-21 10:15:48 +0800 | [diff] [blame] | 564 | |
| 565 | if (!len) |
| 566 | return 0; |
Sage Weil | e98b6fe | 2010-11-09 12:24:53 -0800 | [diff] [blame] | 567 | /* |
| 568 | * flush any page cache pages in this range. this |
| 569 | * will make concurrent normal and sync io slow, |
| 570 | * but it will at least behave sensibly when they are |
| 571 | * in sequence. |
| 572 | */ |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 573 | ret = filemap_write_and_wait_range(inode->i_mapping, off, |
| 574 | off + len); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 575 | if (ret < 0) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 576 | return ret; |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 577 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 578 | num_pages = calc_pages_for(off, len); |
| 579 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
| 580 | if (IS_ERR(pages)) |
| 581 | return PTR_ERR(pages); |
| 582 | ret = striped_read(inode, off, len, pages, |
| 583 | num_pages, checkeof); |
| 584 | if (ret > 0) { |
| 585 | int l, k = 0; |
| 586 | size_t left = ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 587 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 588 | while (left) { |
| 589 | size_t page_off = off & ~PAGE_MASK; |
| 590 | size_t copy = min_t(size_t, left, |
| 591 | PAGE_SIZE - page_off); |
| 592 | l = copy_page_to_iter(pages[k++], page_off, copy, i); |
| 593 | off += l; |
| 594 | left -= l; |
| 595 | if (l < copy) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 596 | break; |
| 597 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 598 | } |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 599 | ceph_release_page_vector(pages, num_pages); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 600 | |
| 601 | if (off > iocb->ki_pos) { |
| 602 | ret = off - iocb->ki_pos; |
| 603 | iocb->ki_pos = off; |
| 604 | } |
| 605 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 606 | dout("sync_read result %d\n", ret); |
| 607 | return ret; |
| 608 | } |
| 609 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 610 | struct ceph_aio_request { |
| 611 | struct kiocb *iocb; |
| 612 | size_t total_len; |
| 613 | int write; |
| 614 | int error; |
| 615 | struct list_head osd_reqs; |
| 616 | unsigned num_reqs; |
| 617 | atomic_t pending_reqs; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 618 | struct timespec mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 619 | struct ceph_cap_flush *prealloc_cf; |
| 620 | }; |
| 621 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 622 | struct ceph_aio_work { |
| 623 | struct work_struct work; |
| 624 | struct ceph_osd_request *req; |
| 625 | }; |
| 626 | |
| 627 | static void ceph_aio_retry_work(struct work_struct *work); |
| 628 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 629 | static void ceph_aio_complete(struct inode *inode, |
| 630 | struct ceph_aio_request *aio_req) |
| 631 | { |
| 632 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 633 | int ret; |
| 634 | |
| 635 | if (!atomic_dec_and_test(&aio_req->pending_reqs)) |
| 636 | return; |
| 637 | |
| 638 | ret = aio_req->error; |
| 639 | if (!ret) |
| 640 | ret = aio_req->total_len; |
| 641 | |
| 642 | dout("ceph_aio_complete %p rc %d\n", inode, ret); |
| 643 | |
| 644 | if (ret >= 0 && aio_req->write) { |
| 645 | int dirty; |
| 646 | |
| 647 | loff_t endoff = aio_req->iocb->ki_pos + aio_req->total_len; |
| 648 | if (endoff > i_size_read(inode)) { |
| 649 | if (ceph_inode_set_size(inode, endoff)) |
| 650 | ceph_check_caps(ci, CHECK_CAPS_AUTHONLY, NULL); |
| 651 | } |
| 652 | |
| 653 | spin_lock(&ci->i_ceph_lock); |
| 654 | ci->i_inline_version = CEPH_INLINE_NONE; |
| 655 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 656 | &aio_req->prealloc_cf); |
| 657 | spin_unlock(&ci->i_ceph_lock); |
| 658 | if (dirty) |
| 659 | __mark_inode_dirty(inode, dirty); |
| 660 | |
| 661 | } |
| 662 | |
| 663 | ceph_put_cap_refs(ci, (aio_req->write ? CEPH_CAP_FILE_WR : |
| 664 | CEPH_CAP_FILE_RD)); |
| 665 | |
| 666 | aio_req->iocb->ki_complete(aio_req->iocb, ret, 0); |
| 667 | |
| 668 | ceph_free_cap_flush(aio_req->prealloc_cf); |
| 669 | kfree(aio_req); |
| 670 | } |
| 671 | |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 672 | static void ceph_aio_complete_req(struct ceph_osd_request *req) |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 673 | { |
| 674 | int rc = req->r_result; |
| 675 | struct inode *inode = req->r_inode; |
| 676 | struct ceph_aio_request *aio_req = req->r_priv; |
| 677 | struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0); |
| 678 | int num_pages = calc_pages_for((u64)osd_data->alignment, |
| 679 | osd_data->length); |
| 680 | |
| 681 | dout("ceph_aio_complete_req %p rc %d bytes %llu\n", |
| 682 | inode, rc, osd_data->length); |
| 683 | |
| 684 | if (rc == -EOLDSNAPC) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 685 | struct ceph_aio_work *aio_work; |
| 686 | BUG_ON(!aio_req->write); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 687 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 688 | aio_work = kmalloc(sizeof(*aio_work), GFP_NOFS); |
| 689 | if (aio_work) { |
| 690 | INIT_WORK(&aio_work->work, ceph_aio_retry_work); |
| 691 | aio_work->req = req; |
| 692 | queue_work(ceph_inode_to_client(inode)->wb_wq, |
| 693 | &aio_work->work); |
| 694 | return; |
| 695 | } |
| 696 | rc = -ENOMEM; |
| 697 | } else if (!aio_req->write) { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 698 | if (rc == -ENOENT) |
| 699 | rc = 0; |
| 700 | if (rc >= 0 && osd_data->length > rc) { |
| 701 | int zoff = osd_data->alignment + rc; |
| 702 | int zlen = osd_data->length - rc; |
| 703 | /* |
| 704 | * If read is satisfied by single OSD request, |
| 705 | * it can pass EOF. Otherwise read is within |
| 706 | * i_size. |
| 707 | */ |
| 708 | if (aio_req->num_reqs == 1) { |
| 709 | loff_t i_size = i_size_read(inode); |
| 710 | loff_t endoff = aio_req->iocb->ki_pos + rc; |
| 711 | if (endoff < i_size) |
| 712 | zlen = min_t(size_t, zlen, |
| 713 | i_size - endoff); |
| 714 | aio_req->total_len = rc + zlen; |
| 715 | } |
| 716 | |
| 717 | if (zlen > 0) |
| 718 | ceph_zero_page_vector_range(zoff, zlen, |
| 719 | osd_data->pages); |
| 720 | } |
| 721 | } |
| 722 | |
| 723 | ceph_put_page_vector(osd_data->pages, num_pages, false); |
| 724 | ceph_osdc_put_request(req); |
| 725 | |
| 726 | if (rc < 0) |
| 727 | cmpxchg(&aio_req->error, 0, rc); |
| 728 | |
| 729 | ceph_aio_complete(inode, aio_req); |
| 730 | return; |
| 731 | } |
| 732 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 733 | static void ceph_aio_retry_work(struct work_struct *work) |
| 734 | { |
| 735 | struct ceph_aio_work *aio_work = |
| 736 | container_of(work, struct ceph_aio_work, work); |
| 737 | struct ceph_osd_request *orig_req = aio_work->req; |
| 738 | struct ceph_aio_request *aio_req = orig_req->r_priv; |
| 739 | struct inode *inode = orig_req->r_inode; |
| 740 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 741 | struct ceph_snap_context *snapc; |
| 742 | struct ceph_osd_request *req; |
| 743 | int ret; |
| 744 | |
| 745 | spin_lock(&ci->i_ceph_lock); |
| 746 | if (__ceph_have_pending_cap_snap(ci)) { |
| 747 | struct ceph_cap_snap *capsnap = |
| 748 | list_last_entry(&ci->i_cap_snaps, |
| 749 | struct ceph_cap_snap, |
| 750 | ci_item); |
| 751 | snapc = ceph_get_snap_context(capsnap->context); |
| 752 | } else { |
| 753 | BUG_ON(!ci->i_head_snapc); |
| 754 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
| 755 | } |
| 756 | spin_unlock(&ci->i_ceph_lock); |
| 757 | |
| 758 | req = ceph_osdc_alloc_request(orig_req->r_osdc, snapc, 2, |
| 759 | false, GFP_NOFS); |
Dan Carpenter | 1418bf0 | 2016-01-26 12:24:44 +0300 | [diff] [blame] | 760 | if (!req) { |
| 761 | ret = -ENOMEM; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 762 | req = orig_req; |
| 763 | goto out; |
| 764 | } |
| 765 | |
| 766 | req->r_flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 767 | CEPH_OSD_FLAG_ONDISK | |
| 768 | CEPH_OSD_FLAG_WRITE; |
Ilya Dryomov | 63244fa | 2016-04-28 16:07:23 +0200 | [diff] [blame] | 769 | ceph_oloc_copy(&req->r_base_oloc, &orig_req->r_base_oloc); |
Ilya Dryomov | d30291b | 2016-04-29 19:54:20 +0200 | [diff] [blame] | 770 | ceph_oid_copy(&req->r_base_oid, &orig_req->r_base_oid); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 771 | |
Ilya Dryomov | 13d1ad1 | 2016-04-27 14:15:51 +0200 | [diff] [blame] | 772 | ret = ceph_osdc_alloc_messages(req, GFP_NOFS); |
| 773 | if (ret) { |
| 774 | ceph_osdc_put_request(req); |
| 775 | req = orig_req; |
| 776 | goto out; |
| 777 | } |
| 778 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 779 | req->r_ops[0] = orig_req->r_ops[0]; |
| 780 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
| 781 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 782 | req->r_mtime = aio_req->mtime; |
| 783 | req->r_data_offset = req->r_ops[0].extent.offset; |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 784 | |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 785 | ceph_osdc_put_request(orig_req); |
| 786 | |
| 787 | req->r_callback = ceph_aio_complete_req; |
| 788 | req->r_inode = inode; |
| 789 | req->r_priv = aio_req; |
| 790 | |
| 791 | ret = ceph_osdc_start_request(req->r_osdc, req, false); |
| 792 | out: |
| 793 | if (ret < 0) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 794 | req->r_result = ret; |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 795 | ceph_aio_complete_req(req); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 796 | } |
| 797 | |
Yan, Zheng | db6aed7 | 2016-01-26 23:05:37 +0800 | [diff] [blame] | 798 | ceph_put_snap_context(snapc); |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 799 | kfree(aio_work); |
| 800 | } |
| 801 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 802 | /* |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 803 | * Write commit request unsafe callback, called to tell us when a |
| 804 | * request is unsafe (that is, in flight--has been handed to the |
| 805 | * messenger to send to its target osd). It is called again when |
| 806 | * we've received a response message indicating the request is |
| 807 | * "safe" (its CEPH_OSD_FLAG_ONDISK flag is set), or when a request |
| 808 | * is completed early (and unsuccessfully) due to a timeout or |
| 809 | * interrupt. |
| 810 | * |
| 811 | * This is used if we requested both an ACK and ONDISK commit reply |
| 812 | * from the OSD. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 813 | */ |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 814 | 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] | 815 | { |
| 816 | struct ceph_inode_info *ci = ceph_inode(req->r_inode); |
| 817 | |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 818 | dout("%s %p tid %llu %ssafe\n", __func__, req, req->r_tid, |
| 819 | unsafe ? "un" : ""); |
| 820 | if (unsafe) { |
| 821 | ceph_get_cap_refs(ci, CEPH_CAP_FILE_WR); |
| 822 | spin_lock(&ci->i_unsafe_lock); |
| 823 | list_add_tail(&req->r_unsafe_item, |
| 824 | &ci->i_unsafe_writes); |
| 825 | spin_unlock(&ci->i_unsafe_lock); |
Ilya Dryomov | fe5da05 | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 826 | |
| 827 | complete_all(&req->r_completion); |
Alex Elder | 26be880 | 2013-04-15 11:20:42 -0500 | [diff] [blame] | 828 | } else { |
| 829 | spin_lock(&ci->i_unsafe_lock); |
| 830 | list_del_init(&req->r_unsafe_item); |
| 831 | spin_unlock(&ci->i_unsafe_lock); |
| 832 | ceph_put_cap_refs(ci, CEPH_CAP_FILE_WR); |
| 833 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 834 | } |
| 835 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 836 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 837 | static ssize_t |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 838 | ceph_direct_read_write(struct kiocb *iocb, struct iov_iter *iter, |
| 839 | struct ceph_snap_context *snapc, |
| 840 | struct ceph_cap_flush **pcf) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 841 | { |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 842 | struct file *file = iocb->ki_filp; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 843 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 844 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 845 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
Alex Elder | acead00 | 2013-03-14 14:09:05 -0500 | [diff] [blame] | 846 | struct ceph_vino vino; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 847 | struct ceph_osd_request *req; |
| 848 | struct page **pages; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 849 | struct ceph_aio_request *aio_req = NULL; |
| 850 | int num_pages = 0; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 851 | int flags; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 852 | int ret; |
Deepa Dinamani | 8bbd471 | 2016-02-02 22:07:48 -0800 | [diff] [blame] | 853 | struct timespec mtime = current_fs_time(inode->i_sb); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 854 | size_t count = iov_iter_count(iter); |
| 855 | loff_t pos = iocb->ki_pos; |
| 856 | bool write = iov_iter_rw(iter) == WRITE; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 857 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 858 | if (write && ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 859 | return -EROFS; |
| 860 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 861 | dout("sync_direct_read_write (%s) on file %p %lld~%u\n", |
| 862 | (write ? "write" : "read"), file, pos, (unsigned)count); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 863 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 864 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 865 | if (ret < 0) |
| 866 | return ret; |
| 867 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 868 | if (write) { |
| 869 | ret = invalidate_inode_pages2_range(inode->i_mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 870 | pos >> PAGE_SHIFT, |
| 871 | (pos + count) >> PAGE_SHIFT); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 872 | if (ret < 0) |
| 873 | dout("invalidate_inode_pages2_range returned %d\n", ret); |
Yehuda Sadeh | 29065a5 | 2010-02-09 11:14:41 -0800 | [diff] [blame] | 874 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 875 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 876 | CEPH_OSD_FLAG_ONDISK | |
| 877 | CEPH_OSD_FLAG_WRITE; |
| 878 | } else { |
| 879 | flags = CEPH_OSD_FLAG_READ; |
| 880 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 881 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 882 | while (iov_iter_count(iter) > 0) { |
| 883 | u64 size = dio_get_pagev_size(iter); |
| 884 | size_t start = 0; |
| 885 | ssize_t len; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 886 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 887 | vino = ceph_vino(inode); |
| 888 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 889 | vino, pos, &size, 0, |
| 890 | /*include a 'startsync' command*/ |
| 891 | write ? 2 : 1, |
| 892 | write ? CEPH_OSD_OP_WRITE : |
| 893 | CEPH_OSD_OP_READ, |
| 894 | flags, snapc, |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 895 | ci->i_truncate_seq, |
| 896 | ci->i_truncate_size, |
| 897 | false); |
| 898 | if (IS_ERR(req)) { |
| 899 | ret = PTR_ERR(req); |
Al Viro | eab8723 | 2014-04-03 22:44:19 -0400 | [diff] [blame] | 900 | break; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 901 | } |
| 902 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 903 | len = size; |
| 904 | pages = dio_get_pages_alloc(iter, len, &start, &num_pages); |
Zhu, Caifeng | b5b9898 | 2015-10-08 15:26:15 +0800 | [diff] [blame] | 905 | if (IS_ERR(pages)) { |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 906 | ceph_osdc_put_request(req); |
Zhu, Caifeng | b5b9898 | 2015-10-08 15:26:15 +0800 | [diff] [blame] | 907 | ret = PTR_ERR(pages); |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 908 | break; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 909 | } |
| 910 | |
| 911 | /* |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 912 | * To simplify error handling, allow AIO when IO within i_size |
| 913 | * or IO can be satisfied by single OSD request. |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 914 | */ |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 915 | if (pos == iocb->ki_pos && !is_sync_kiocb(iocb) && |
| 916 | (len == count || pos + count <= i_size_read(inode))) { |
| 917 | aio_req = kzalloc(sizeof(*aio_req), GFP_KERNEL); |
| 918 | if (aio_req) { |
| 919 | aio_req->iocb = iocb; |
| 920 | aio_req->write = write; |
| 921 | INIT_LIST_HEAD(&aio_req->osd_reqs); |
| 922 | if (write) { |
Yan, Zheng | 5be0389 | 2015-12-24 08:44:20 +0800 | [diff] [blame] | 923 | aio_req->mtime = mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 924 | swap(aio_req->prealloc_cf, *pcf); |
| 925 | } |
| 926 | } |
| 927 | /* ignore error */ |
| 928 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 929 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 930 | if (write) { |
| 931 | /* |
| 932 | * throw out any page cache pages in this range. this |
| 933 | * may block. |
| 934 | */ |
| 935 | truncate_inode_pages_range(inode->i_mapping, pos, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 936 | (pos+len) | (PAGE_SIZE - 1)); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 937 | |
| 938 | osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC, 0); |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 939 | req->r_mtime = mtime; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 940 | } |
| 941 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 942 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, start, |
| 943 | false, false); |
| 944 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 945 | if (aio_req) { |
| 946 | aio_req->total_len += len; |
| 947 | aio_req->num_reqs++; |
| 948 | atomic_inc(&aio_req->pending_reqs); |
| 949 | |
| 950 | req->r_callback = ceph_aio_complete_req; |
| 951 | req->r_inode = inode; |
| 952 | req->r_priv = aio_req; |
| 953 | list_add_tail(&req->r_unsafe_item, &aio_req->osd_reqs); |
| 954 | |
| 955 | pos += len; |
| 956 | iov_iter_advance(iter, len); |
| 957 | continue; |
| 958 | } |
| 959 | |
| 960 | ret = ceph_osdc_start_request(req->r_osdc, req, false); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 961 | if (!ret) |
| 962 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
| 963 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 964 | size = i_size_read(inode); |
| 965 | if (!write) { |
| 966 | if (ret == -ENOENT) |
| 967 | ret = 0; |
| 968 | if (ret >= 0 && ret < len && pos + ret < size) { |
| 969 | int zlen = min_t(size_t, len - ret, |
| 970 | size - pos - ret); |
| 971 | ceph_zero_page_vector_range(start + ret, zlen, |
| 972 | pages); |
| 973 | ret += zlen; |
| 974 | } |
| 975 | if (ret >= 0) |
| 976 | len = ret; |
| 977 | } |
| 978 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 979 | ceph_put_page_vector(pages, num_pages, false); |
| 980 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 981 | ceph_osdc_put_request(req); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 982 | if (ret < 0) |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 983 | break; |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 984 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 985 | pos += len; |
| 986 | iov_iter_advance(iter, len); |
| 987 | |
| 988 | if (!write && pos >= size) |
| 989 | break; |
| 990 | |
| 991 | if (write && pos > size) { |
| 992 | if (ceph_inode_set_size(inode, pos)) |
Al Viro | 64c3131 | 2014-04-03 22:58:25 -0400 | [diff] [blame] | 993 | ceph_check_caps(ceph_inode(inode), |
| 994 | CHECK_CAPS_AUTHONLY, |
| 995 | NULL); |
| 996 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 997 | } |
| 998 | |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 999 | if (aio_req) { |
| 1000 | if (aio_req->num_reqs == 0) { |
| 1001 | kfree(aio_req); |
| 1002 | return ret; |
| 1003 | } |
| 1004 | |
| 1005 | ceph_get_cap_refs(ci, write ? CEPH_CAP_FILE_WR : |
| 1006 | CEPH_CAP_FILE_RD); |
| 1007 | |
| 1008 | while (!list_empty(&aio_req->osd_reqs)) { |
| 1009 | req = list_first_entry(&aio_req->osd_reqs, |
| 1010 | struct ceph_osd_request, |
| 1011 | r_unsafe_item); |
| 1012 | list_del_init(&req->r_unsafe_item); |
| 1013 | if (ret >= 0) |
| 1014 | ret = ceph_osdc_start_request(req->r_osdc, |
| 1015 | req, false); |
| 1016 | if (ret < 0) { |
| 1017 | req->r_result = ret; |
Ilya Dryomov | 85e084f | 2016-04-28 16:07:24 +0200 | [diff] [blame] | 1018 | ceph_aio_complete_req(req); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1019 | } |
| 1020 | } |
| 1021 | return -EIOCBQUEUED; |
| 1022 | } |
| 1023 | |
| 1024 | if (ret != -EOLDSNAPC && pos > iocb->ki_pos) { |
| 1025 | ret = pos - iocb->ki_pos; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1026 | iocb->ki_pos = pos; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1027 | } |
| 1028 | return ret; |
| 1029 | } |
| 1030 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1031 | /* |
| 1032 | * Synchronous write, straight from __user pointer or user pages. |
| 1033 | * |
| 1034 | * If write spans object boundary, just do multiple writes. (For a |
| 1035 | * correct atomic write, we should e.g. take write locks on all |
| 1036 | * objects, rollback on failure, etc.) |
| 1037 | */ |
Yan, Zheng | 06fee30 | 2014-07-28 14:33:46 +0800 | [diff] [blame] | 1038 | static ssize_t |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1039 | ceph_sync_write(struct kiocb *iocb, struct iov_iter *from, loff_t pos, |
| 1040 | struct ceph_snap_context *snapc) |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1041 | { |
| 1042 | struct file *file = iocb->ki_filp; |
| 1043 | struct inode *inode = file_inode(file); |
| 1044 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1045 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1046 | struct ceph_vino vino; |
| 1047 | struct ceph_osd_request *req; |
| 1048 | struct page **pages; |
| 1049 | u64 len; |
| 1050 | int num_pages; |
| 1051 | int written = 0; |
| 1052 | int flags; |
| 1053 | int check_caps = 0; |
| 1054 | int ret; |
Deepa Dinamani | 8bbd471 | 2016-02-02 22:07:48 -0800 | [diff] [blame] | 1055 | struct timespec mtime = current_fs_time(inode->i_sb); |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1056 | size_t count = iov_iter_count(from); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1057 | |
| 1058 | if (ceph_snap(file_inode(file)) != CEPH_NOSNAP) |
| 1059 | return -EROFS; |
| 1060 | |
| 1061 | dout("sync_write on file %p %lld~%u\n", file, pos, (unsigned)count); |
| 1062 | |
| 1063 | ret = filemap_write_and_wait_range(inode->i_mapping, pos, pos + count); |
| 1064 | if (ret < 0) |
| 1065 | return ret; |
| 1066 | |
| 1067 | ret = invalidate_inode_pages2_range(inode->i_mapping, |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1068 | pos >> PAGE_SHIFT, |
| 1069 | (pos + count) >> PAGE_SHIFT); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1070 | if (ret < 0) |
| 1071 | dout("invalidate_inode_pages2_range returned %d\n", ret); |
| 1072 | |
| 1073 | flags = CEPH_OSD_FLAG_ORDERSNAP | |
| 1074 | CEPH_OSD_FLAG_ONDISK | |
| 1075 | CEPH_OSD_FLAG_WRITE | |
| 1076 | CEPH_OSD_FLAG_ACK; |
| 1077 | |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1078 | while ((len = iov_iter_count(from)) > 0) { |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1079 | size_t left; |
| 1080 | int n; |
| 1081 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1082 | vino = ceph_vino(inode); |
| 1083 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
Yan, Zheng | 715e4cd | 2014-11-13 14:40:37 +0800 | [diff] [blame] | 1084 | vino, pos, &len, 0, 1, |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1085 | CEPH_OSD_OP_WRITE, flags, snapc, |
| 1086 | ci->i_truncate_seq, |
| 1087 | ci->i_truncate_size, |
| 1088 | false); |
| 1089 | if (IS_ERR(req)) { |
| 1090 | ret = PTR_ERR(req); |
Al Viro | eab8723 | 2014-04-03 22:44:19 -0400 | [diff] [blame] | 1091 | break; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1092 | } |
| 1093 | |
| 1094 | /* |
| 1095 | * write from beginning of first page, |
| 1096 | * regardless of io alignment |
| 1097 | */ |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1098 | num_pages = (len + PAGE_SIZE - 1) >> PAGE_SHIFT; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1099 | |
Yan, Zheng | 687265e | 2015-06-13 17:27:05 +0800 | [diff] [blame] | 1100 | pages = ceph_alloc_page_vector(num_pages, GFP_KERNEL); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1101 | if (IS_ERR(pages)) { |
| 1102 | ret = PTR_ERR(pages); |
| 1103 | goto out; |
| 1104 | } |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1105 | |
| 1106 | left = len; |
| 1107 | for (n = 0; n < num_pages; n++) { |
Ilya Dryomov | 125d725 | 2014-01-28 19:16:18 +0200 | [diff] [blame] | 1108 | size_t plen = min_t(size_t, left, PAGE_SIZE); |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1109 | ret = copy_page_from_iter(pages[n], 0, plen, from); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1110 | if (ret != plen) { |
| 1111 | ret = -EFAULT; |
| 1112 | break; |
| 1113 | } |
| 1114 | left -= ret; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1115 | } |
| 1116 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1117 | if (ret < 0) { |
| 1118 | ceph_release_page_vector(pages, num_pages); |
| 1119 | goto out; |
| 1120 | } |
| 1121 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1122 | /* get a second commit callback */ |
| 1123 | req->r_unsafe_callback = ceph_sync_write_unsafe; |
| 1124 | req->r_inode = inode; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1125 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1126 | osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, |
| 1127 | false, true); |
Alex Elder | 02ee07d | 2013-03-14 14:09:06 -0500 | [diff] [blame] | 1128 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 1129 | req->r_mtime = mtime; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1130 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
| 1131 | if (!ret) |
| 1132 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1133 | |
| 1134 | out: |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1135 | ceph_osdc_put_request(req); |
| 1136 | if (ret == 0) { |
| 1137 | pos += len; |
| 1138 | written += len; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1139 | |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1140 | if (pos > i_size_read(inode)) { |
| 1141 | check_caps = ceph_inode_set_size(inode, pos); |
| 1142 | if (check_caps) |
| 1143 | ceph_check_caps(ceph_inode(inode), |
| 1144 | CHECK_CAPS_AUTHONLY, |
| 1145 | NULL); |
| 1146 | } |
| 1147 | } else |
| 1148 | break; |
| 1149 | } |
| 1150 | |
| 1151 | if (ret != -EOLDSNAPC && written > 0) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1152 | ret = written; |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1153 | iocb->ki_pos = pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1154 | } |
| 1155 | return ret; |
| 1156 | } |
| 1157 | |
| 1158 | /* |
| 1159 | * Wrap generic_file_aio_read with checks for cap bits on the inode. |
| 1160 | * Atomically grab references, so that those bits are not released |
| 1161 | * back to the MDS mid-read. |
| 1162 | * |
| 1163 | * Hmm, the sync read case isn't actually async... should it be? |
| 1164 | */ |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1165 | 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] | 1166 | { |
| 1167 | struct file *filp = iocb->ki_filp; |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1168 | struct ceph_file_info *fi = filp->private_data; |
Christoph Hellwig | 66ee59a | 2015-02-11 19:56:46 +0100 | [diff] [blame] | 1169 | size_t len = iov_iter_count(to); |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1170 | struct inode *inode = file_inode(filp); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1171 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1172 | struct page *pinned_page = NULL; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1173 | ssize_t ret; |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1174 | int want, got = 0; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1175 | int retry_op = 0, read = 0; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1176 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1177 | again: |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1178 | dout("aio_read %p %llx.%llx %llu~%u trying to get caps on %p\n", |
| 1179 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, inode); |
| 1180 | |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1181 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1182 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; |
| 1183 | else |
| 1184 | want = CEPH_CAP_FILE_CACHE; |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1185 | 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] | 1186 | if (ret < 0) |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1187 | return ret; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1188 | |
Sage Weil | 2962507 | 2010-05-27 10:40:43 -0700 | [diff] [blame] | 1189 | if ((got & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0 || |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1190 | (iocb->ki_flags & IOCB_DIRECT) || |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1191 | (fi->flags & CEPH_F_SYNC)) { |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1192 | |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1193 | dout("aio_sync_read %p %llx.%llx %llu~%u got cap refs on %s\n", |
| 1194 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
| 1195 | ceph_cap_string(got)); |
| 1196 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1197 | if (ci->i_inline_version == CEPH_INLINE_NONE) { |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1198 | if (!retry_op && (iocb->ki_flags & IOCB_DIRECT)) { |
| 1199 | ret = ceph_direct_read_write(iocb, to, |
| 1200 | NULL, NULL); |
| 1201 | if (ret >= 0 && ret < len) |
| 1202 | retry_op = CHECK_EOF; |
| 1203 | } else { |
| 1204 | ret = ceph_sync_read(iocb, to, &retry_op); |
| 1205 | } |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1206 | } else { |
| 1207 | retry_op = READ_INLINE; |
| 1208 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1209 | } else { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1210 | 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] | 1211 | inode, ceph_vinop(inode), iocb->ki_pos, (unsigned)len, |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1212 | ceph_cap_string(got)); |
| 1213 | |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1214 | ret = generic_file_read_iter(iocb, to); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1215 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1216 | dout("aio_read %p %llx.%llx dropping cap refs on %s = %d\n", |
| 1217 | inode, ceph_vinop(inode), ceph_cap_string(got), (int)ret); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1218 | if (pinned_page) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1219 | put_page(pinned_page); |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1220 | pinned_page = NULL; |
| 1221 | } |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1222 | ceph_put_cap_refs(ci, got); |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1223 | if (retry_op > HAVE_RETRIED && ret >= 0) { |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1224 | int statret; |
| 1225 | struct page *page = NULL; |
| 1226 | loff_t i_size; |
| 1227 | if (retry_op == READ_INLINE) { |
Yan, Zheng | 687265e | 2015-06-13 17:27:05 +0800 | [diff] [blame] | 1228 | page = __page_cache_alloc(GFP_KERNEL); |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1229 | if (!page) |
| 1230 | return -ENOMEM; |
| 1231 | } |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1232 | |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1233 | statret = __ceph_do_getattr(inode, page, |
| 1234 | CEPH_STAT_CAP_INLINE_DATA, !!page); |
| 1235 | if (statret < 0) { |
| 1236 | __free_page(page); |
| 1237 | if (statret == -ENODATA) { |
| 1238 | BUG_ON(retry_op != READ_INLINE); |
| 1239 | goto again; |
| 1240 | } |
| 1241 | return statret; |
| 1242 | } |
| 1243 | |
| 1244 | i_size = i_size_read(inode); |
| 1245 | if (retry_op == READ_INLINE) { |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1246 | BUG_ON(ret > 0 || read > 0); |
| 1247 | if (iocb->ki_pos < i_size && |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1248 | iocb->ki_pos < PAGE_SIZE) { |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1249 | loff_t end = min_t(loff_t, i_size, |
| 1250 | iocb->ki_pos + len); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1251 | end = min_t(loff_t, end, PAGE_SIZE); |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1252 | if (statret < end) |
| 1253 | zero_user_segment(page, statret, end); |
| 1254 | ret = copy_page_to_iter(page, |
| 1255 | iocb->ki_pos & ~PAGE_MASK, |
| 1256 | end - iocb->ki_pos, to); |
| 1257 | iocb->ki_pos += ret; |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1258 | read += ret; |
| 1259 | } |
| 1260 | if (iocb->ki_pos < i_size && read < len) { |
| 1261 | size_t zlen = min_t(size_t, len - read, |
| 1262 | i_size - iocb->ki_pos); |
| 1263 | ret = iov_iter_zero(zlen, to); |
| 1264 | iocb->ki_pos += ret; |
| 1265 | read += ret; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1266 | } |
| 1267 | __free_pages(page, 0); |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1268 | return read; |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1269 | } |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1270 | |
| 1271 | /* hit EOF or hole? */ |
Yan, Zheng | 8370124 | 2014-11-14 22:36:18 +0800 | [diff] [blame] | 1272 | if (retry_op == CHECK_EOF && iocb->ki_pos < i_size && |
Yan, Zheng | fcc02d2 | 2015-01-10 11:43:12 +0800 | [diff] [blame] | 1273 | ret < len) { |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1274 | dout("sync_read hit hole, ppos %lld < size %lld" |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1275 | ", reading more\n", iocb->ki_pos, i_size); |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1276 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1277 | read += ret; |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1278 | len -= ret; |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1279 | retry_op = HAVE_RETRIED; |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1280 | goto again; |
| 1281 | } |
| 1282 | } |
majianpeng | 8eb4efb | 2013-09-26 14:42:17 +0800 | [diff] [blame] | 1283 | |
Sage Weil | 6a02658 | 2010-02-09 14:04:02 -0800 | [diff] [blame] | 1284 | if (ret >= 0) |
| 1285 | ret += read; |
| 1286 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1287 | return ret; |
| 1288 | } |
| 1289 | |
| 1290 | /* |
| 1291 | * Take cap references to avoid releasing caps to MDS mid-write. |
| 1292 | * |
| 1293 | * If we are synchronous, and write with an old snap context, the OSD |
| 1294 | * may return EOLDSNAPC. In that case, retry the write.. _after_ |
| 1295 | * dropping our cap refs and allowing the pending snap to logically |
| 1296 | * complete _before_ this write occurs. |
| 1297 | * |
| 1298 | * If we are near ENOSPC, write synchronously. |
| 1299 | */ |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1300 | 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] | 1301 | { |
| 1302 | struct file *file = iocb->ki_filp; |
Sage Weil | 33caad3 | 2010-05-26 14:31:27 -0700 | [diff] [blame] | 1303 | struct ceph_file_info *fi = file->private_data; |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 1304 | struct inode *inode = file_inode(file); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1305 | struct ceph_inode_info *ci = ceph_inode(inode); |
Yehuda Sadeh | 3d14c5d | 2010-04-06 15:14:15 -0700 | [diff] [blame] | 1306 | struct ceph_osd_client *osdc = |
| 1307 | &ceph_sb_to_client(inode->i_sb)->client->osdc; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1308 | struct ceph_cap_flush *prealloc_cf; |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1309 | ssize_t count, written = 0; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1310 | int err, want, got; |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1311 | loff_t pos; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1312 | |
| 1313 | if (ceph_snap(inode) != CEPH_NOSNAP) |
| 1314 | return -EROFS; |
| 1315 | |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1316 | prealloc_cf = ceph_alloc_cap_flush(); |
| 1317 | if (!prealloc_cf) |
| 1318 | return -ENOMEM; |
| 1319 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1320 | inode_lock(inode); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1321 | |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1322 | /* We can write back this queue in page reclaim */ |
Christoph Hellwig | de1414a | 2015-01-14 10:42:36 +0100 | [diff] [blame] | 1323 | current->backing_dev_info = inode_to_bdi(inode); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1324 | |
Yan, Zheng | 55b0b31 | 2015-09-07 11:35:01 +0800 | [diff] [blame] | 1325 | if (iocb->ki_flags & IOCB_APPEND) { |
| 1326 | err = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
| 1327 | if (err < 0) |
| 1328 | goto out; |
| 1329 | } |
| 1330 | |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1331 | err = generic_write_checks(iocb, from); |
| 1332 | if (err <= 0) |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1333 | goto out; |
| 1334 | |
Al Viro | 3309dd0 | 2015-04-09 12:55:47 -0400 | [diff] [blame] | 1335 | pos = iocb->ki_pos; |
| 1336 | count = iov_iter_count(from); |
Jan Kara | 5fa8e0a | 2015-05-21 16:05:53 +0200 | [diff] [blame] | 1337 | err = file_remove_privs(file); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1338 | if (err) |
| 1339 | goto out; |
| 1340 | |
| 1341 | err = file_update_time(file); |
| 1342 | if (err) |
| 1343 | goto out; |
| 1344 | |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1345 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
| 1346 | err = ceph_uninline_data(file, NULL); |
| 1347 | if (err < 0) |
| 1348 | goto out; |
| 1349 | } |
| 1350 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1351 | retry_snap: |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1352 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL)) { |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1353 | err = -ENOSPC; |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1354 | goto out; |
| 1355 | } |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1356 | |
Randy Dunlap | ac7f29b | 2013-04-19 14:20:07 -0700 | [diff] [blame] | 1357 | 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] | 1358 | inode, ceph_vinop(inode), pos, count, i_size_read(inode)); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1359 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1360 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; |
| 1361 | else |
| 1362 | want = CEPH_CAP_FILE_BUFFER; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1363 | got = 0; |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1364 | err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, pos + count, |
| 1365 | &got, NULL); |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1366 | if (err < 0) |
Yan, Zheng | 37505d5 | 2013-04-12 16:11:10 +0800 | [diff] [blame] | 1367 | goto out; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1368 | |
Randy Dunlap | ac7f29b | 2013-04-19 14:20:07 -0700 | [diff] [blame] | 1369 | 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] | 1370 | inode, ceph_vinop(inode), pos, count, ceph_cap_string(got)); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1371 | |
| 1372 | if ((got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO)) == 0 || |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1373 | (iocb->ki_flags & IOCB_DIRECT) || (fi->flags & CEPH_F_SYNC)) { |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1374 | struct ceph_snap_context *snapc; |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1375 | struct iov_iter data; |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1376 | inode_unlock(inode); |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1377 | |
| 1378 | spin_lock(&ci->i_ceph_lock); |
| 1379 | if (__ceph_have_pending_cap_snap(ci)) { |
| 1380 | struct ceph_cap_snap *capsnap = |
| 1381 | list_last_entry(&ci->i_cap_snaps, |
| 1382 | struct ceph_cap_snap, |
| 1383 | ci_item); |
| 1384 | snapc = ceph_get_snap_context(capsnap->context); |
| 1385 | } else { |
| 1386 | BUG_ON(!ci->i_head_snapc); |
| 1387 | snapc = ceph_get_snap_context(ci->i_head_snapc); |
| 1388 | } |
| 1389 | spin_unlock(&ci->i_ceph_lock); |
| 1390 | |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1391 | /* we might need to revert back to that point */ |
| 1392 | data = *from; |
Al Viro | 2ba48ce | 2015-04-09 13:52:01 -0400 | [diff] [blame] | 1393 | if (iocb->ki_flags & IOCB_DIRECT) |
Yan, Zheng | c8fe9b1 | 2015-12-23 21:23:38 +0800 | [diff] [blame] | 1394 | written = ceph_direct_read_write(iocb, &data, snapc, |
| 1395 | &prealloc_cf); |
majianpeng | e8344e6 | 2013-09-12 13:54:26 +0800 | [diff] [blame] | 1396 | else |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1397 | written = ceph_sync_write(iocb, &data, pos, snapc); |
majianpeng | 0e5dd45 | 2013-08-08 15:32:19 +0800 | [diff] [blame] | 1398 | if (written == -EOLDSNAPC) { |
| 1399 | dout("aio_write %p %llx.%llx %llu~%u" |
| 1400 | "got EOLDSNAPC, retrying\n", |
| 1401 | inode, ceph_vinop(inode), |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1402 | pos, (unsigned)count); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1403 | inode_lock(inode); |
majianpeng | 0e5dd45 | 2013-08-08 15:32:19 +0800 | [diff] [blame] | 1404 | goto retry_snap; |
| 1405 | } |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1406 | if (written > 0) |
| 1407 | iov_iter_advance(from, written); |
Yan, Zheng | 5dda377c | 2015-04-30 14:40:54 +0800 | [diff] [blame] | 1408 | ceph_put_snap_context(snapc); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1409 | } else { |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1410 | loff_t old_size = i_size_read(inode); |
Yan, Zheng | b0d7c22 | 2013-08-12 21:42:15 -0700 | [diff] [blame] | 1411 | /* |
| 1412 | * No need to acquire the i_truncate_mutex. Because |
| 1413 | * the MDS revokes Fwb caps before sending truncate |
| 1414 | * message to us. We can't get Fwb cap while there |
| 1415 | * are pending vmtruncate. So write and vmtruncate |
| 1416 | * can not run at the same time |
| 1417 | */ |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1418 | written = generic_perform_write(file, from, pos); |
Al Viro | aec605f4 | 2014-02-11 22:28:43 -0500 | [diff] [blame] | 1419 | if (likely(written >= 0)) |
| 1420 | iocb->ki_pos = pos + written; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1421 | if (i_size_read(inode) > old_size) |
Yunchuan Wen | 32d3e14 | 2013-12-26 06:29:27 -0800 | [diff] [blame] | 1422 | ceph_fscache_update_objectsize(inode); |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1423 | inode_unlock(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1424 | } |
Sage Weil | d8de9ab | 2011-07-26 11:27:34 -0700 | [diff] [blame] | 1425 | |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1426 | if (written >= 0) { |
Sage Weil | fca65b4 | 2011-05-04 11:33:47 -0700 | [diff] [blame] | 1427 | int dirty; |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 1428 | spin_lock(&ci->i_ceph_lock); |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1429 | ci->i_inline_version = CEPH_INLINE_NONE; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1430 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 1431 | &prealloc_cf); |
Sage Weil | be65559 | 2011-11-30 09:47:09 -0800 | [diff] [blame] | 1432 | spin_unlock(&ci->i_ceph_lock); |
Sage Weil | fca65b4 | 2011-05-04 11:33:47 -0700 | [diff] [blame] | 1433 | if (dirty) |
| 1434 | __mark_inode_dirty(inode, dirty); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1435 | } |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1436 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1437 | 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] | 1438 | inode, ceph_vinop(inode), pos, (unsigned)count, |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1439 | ceph_cap_string(got)); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1440 | ceph_put_cap_refs(ci, got); |
Sage Weil | 7971bd9 | 2013-05-01 21:15:58 -0700 | [diff] [blame] | 1441 | |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1442 | if (written >= 0 && |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1443 | ((file->f_flags & O_SYNC) || IS_SYNC(file->f_mapping->host) || |
| 1444 | ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_NEARFULL))) { |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1445 | err = vfs_fsync_range(file, pos, pos + written - 1, 1); |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1446 | if (err < 0) |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1447 | written = err; |
Yan, Zheng | 6070e0c | 2013-03-01 10:55:39 +0800 | [diff] [blame] | 1448 | } |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1449 | |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1450 | goto out_unlocked; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1451 | |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1452 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1453 | inode_unlock(inode); |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1454 | out_unlocked: |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1455 | ceph_free_cap_flush(prealloc_cf); |
Sage Weil | 2f75e9e | 2013-08-09 09:57:58 -0700 | [diff] [blame] | 1456 | current->backing_dev_info = NULL; |
Yan, Zheng | 03d254e | 2013-04-12 16:11:13 +0800 | [diff] [blame] | 1457 | return written ? written : err; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1458 | } |
| 1459 | |
| 1460 | /* |
| 1461 | * llseek. be sure to verify file size on SEEK_END. |
| 1462 | */ |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1463 | 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] | 1464 | { |
| 1465 | struct inode *inode = file->f_mapping->host; |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1466 | loff_t i_size; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1467 | int ret; |
| 1468 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1469 | inode_lock(inode); |
Sage Weil | 6a82c47 | 2011-12-13 09:19:26 -0800 | [diff] [blame] | 1470 | |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1471 | if (whence == SEEK_END || whence == SEEK_DATA || whence == SEEK_HOLE) { |
Yan, Zheng | 508b32d | 2014-09-16 21:46:17 +0800 | [diff] [blame] | 1472 | ret = ceph_do_getattr(inode, CEPH_STAT_CAP_SIZE, false); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1473 | if (ret < 0) { |
| 1474 | offset = ret; |
| 1475 | goto out; |
| 1476 | } |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1477 | } |
| 1478 | |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1479 | i_size = i_size_read(inode); |
Andrew Morton | 965c8e5 | 2012-12-17 15:59:39 -0800 | [diff] [blame] | 1480 | switch (whence) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1481 | case SEEK_END: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1482 | offset += i_size; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1483 | break; |
| 1484 | case SEEK_CUR: |
| 1485 | /* |
| 1486 | * Here we special-case the lseek(fd, 0, SEEK_CUR) |
| 1487 | * position-querying operation. Avoid rewriting the "same" |
| 1488 | * f_pos value back to the file because a concurrent read(), |
| 1489 | * write() or lseek() might have altered it |
| 1490 | */ |
| 1491 | if (offset == 0) { |
| 1492 | offset = file->f_pos; |
| 1493 | goto out; |
| 1494 | } |
| 1495 | offset += file->f_pos; |
| 1496 | break; |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1497 | case SEEK_DATA: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1498 | if (offset >= i_size) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1499 | ret = -ENXIO; |
| 1500 | goto out; |
| 1501 | } |
| 1502 | break; |
| 1503 | case SEEK_HOLE: |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1504 | if (offset >= i_size) { |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1505 | ret = -ENXIO; |
| 1506 | goto out; |
| 1507 | } |
Yan, Zheng | 99c88e6 | 2015-12-30 11:32:46 +0800 | [diff] [blame] | 1508 | offset = i_size; |
Josef Bacik | 06222e4 | 2011-07-18 13:21:38 -0400 | [diff] [blame] | 1509 | break; |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1510 | } |
| 1511 | |
Jie Liu | 46a1c2c | 2013-06-25 12:02:13 +0800 | [diff] [blame] | 1512 | offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1513 | |
| 1514 | out: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1515 | inode_unlock(inode); |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1516 | return offset; |
| 1517 | } |
| 1518 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1519 | static inline void ceph_zero_partial_page( |
| 1520 | struct inode *inode, loff_t offset, unsigned size) |
| 1521 | { |
| 1522 | struct page *page; |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1523 | pgoff_t index = offset >> PAGE_SHIFT; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1524 | |
| 1525 | page = find_lock_page(inode->i_mapping, index); |
| 1526 | if (page) { |
| 1527 | wait_on_page_writeback(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1528 | zero_user(page, offset & (PAGE_SIZE - 1), size); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1529 | unlock_page(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1530 | put_page(page); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1531 | } |
| 1532 | } |
| 1533 | |
| 1534 | static void ceph_zero_pagecache_range(struct inode *inode, loff_t offset, |
| 1535 | loff_t length) |
| 1536 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1537 | loff_t nearly = round_up(offset, PAGE_SIZE); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1538 | if (offset < nearly) { |
| 1539 | loff_t size = nearly - offset; |
| 1540 | if (length < size) |
| 1541 | size = length; |
| 1542 | ceph_zero_partial_page(inode, offset, size); |
| 1543 | offset += size; |
| 1544 | length -= size; |
| 1545 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 1546 | if (length >= PAGE_SIZE) { |
| 1547 | loff_t size = round_down(length, PAGE_SIZE); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1548 | truncate_pagecache_range(inode, offset, offset + size - 1); |
| 1549 | offset += size; |
| 1550 | length -= size; |
| 1551 | } |
| 1552 | if (length) |
| 1553 | ceph_zero_partial_page(inode, offset, length); |
| 1554 | } |
| 1555 | |
| 1556 | static int ceph_zero_partial_object(struct inode *inode, |
| 1557 | loff_t offset, loff_t *length) |
| 1558 | { |
| 1559 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1560 | struct ceph_fs_client *fsc = ceph_inode_to_client(inode); |
| 1561 | struct ceph_osd_request *req; |
| 1562 | int ret = 0; |
| 1563 | loff_t zero = 0; |
| 1564 | int op; |
| 1565 | |
| 1566 | if (!length) { |
| 1567 | op = offset ? CEPH_OSD_OP_DELETE : CEPH_OSD_OP_TRUNCATE; |
| 1568 | length = &zero; |
| 1569 | } else { |
| 1570 | op = CEPH_OSD_OP_ZERO; |
| 1571 | } |
| 1572 | |
| 1573 | req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, |
| 1574 | ceph_vino(inode), |
| 1575 | offset, length, |
Yan, Zheng | 715e4cd | 2014-11-13 14:40:37 +0800 | [diff] [blame] | 1576 | 0, 1, op, |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1577 | CEPH_OSD_FLAG_WRITE | |
| 1578 | CEPH_OSD_FLAG_ONDISK, |
| 1579 | NULL, 0, 0, false); |
| 1580 | if (IS_ERR(req)) { |
| 1581 | ret = PTR_ERR(req); |
| 1582 | goto out; |
| 1583 | } |
| 1584 | |
Ilya Dryomov | bb873b5 | 2016-05-26 00:29:52 +0200 | [diff] [blame] | 1585 | req->r_mtime = inode->i_mtime; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1586 | ret = ceph_osdc_start_request(&fsc->client->osdc, req, false); |
| 1587 | if (!ret) { |
| 1588 | ret = ceph_osdc_wait_request(&fsc->client->osdc, req); |
| 1589 | if (ret == -ENOENT) |
| 1590 | ret = 0; |
| 1591 | } |
| 1592 | ceph_osdc_put_request(req); |
| 1593 | |
| 1594 | out: |
| 1595 | return ret; |
| 1596 | } |
| 1597 | |
| 1598 | static int ceph_zero_objects(struct inode *inode, loff_t offset, loff_t length) |
| 1599 | { |
| 1600 | int ret = 0; |
| 1601 | struct ceph_inode_info *ci = ceph_inode(inode); |
Sage Weil | b314a90 | 2013-08-27 12:15:16 -0700 | [diff] [blame] | 1602 | s32 stripe_unit = ceph_file_layout_su(ci->i_layout); |
| 1603 | s32 stripe_count = ceph_file_layout_stripe_count(ci->i_layout); |
| 1604 | s32 object_size = ceph_file_layout_object_size(ci->i_layout); |
| 1605 | u64 object_set_size = object_size * stripe_count; |
| 1606 | u64 nearly, t; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1607 | |
Sage Weil | b314a90 | 2013-08-27 12:15:16 -0700 | [diff] [blame] | 1608 | /* round offset up to next period boundary */ |
| 1609 | nearly = offset + object_set_size - 1; |
| 1610 | t = nearly; |
| 1611 | nearly -= do_div(t, object_set_size); |
| 1612 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1613 | while (length && offset < nearly) { |
| 1614 | loff_t size = length; |
| 1615 | ret = ceph_zero_partial_object(inode, offset, &size); |
| 1616 | if (ret < 0) |
| 1617 | return ret; |
| 1618 | offset += size; |
| 1619 | length -= size; |
| 1620 | } |
| 1621 | while (length >= object_set_size) { |
| 1622 | int i; |
| 1623 | loff_t pos = offset; |
| 1624 | for (i = 0; i < stripe_count; ++i) { |
| 1625 | ret = ceph_zero_partial_object(inode, pos, NULL); |
| 1626 | if (ret < 0) |
| 1627 | return ret; |
| 1628 | pos += stripe_unit; |
| 1629 | } |
| 1630 | offset += object_set_size; |
| 1631 | length -= object_set_size; |
| 1632 | } |
| 1633 | while (length) { |
| 1634 | loff_t size = length; |
| 1635 | ret = ceph_zero_partial_object(inode, offset, &size); |
| 1636 | if (ret < 0) |
| 1637 | return ret; |
| 1638 | offset += size; |
| 1639 | length -= size; |
| 1640 | } |
| 1641 | return ret; |
| 1642 | } |
| 1643 | |
| 1644 | static long ceph_fallocate(struct file *file, int mode, |
| 1645 | loff_t offset, loff_t length) |
| 1646 | { |
| 1647 | struct ceph_file_info *fi = file->private_data; |
Libo Chen | aa8b60e | 2013-12-11 13:49:11 +0800 | [diff] [blame] | 1648 | struct inode *inode = file_inode(file); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1649 | struct ceph_inode_info *ci = ceph_inode(inode); |
| 1650 | struct ceph_osd_client *osdc = |
| 1651 | &ceph_inode_to_client(inode)->client->osdc; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1652 | struct ceph_cap_flush *prealloc_cf; |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1653 | int want, got = 0; |
| 1654 | int dirty; |
| 1655 | int ret = 0; |
| 1656 | loff_t endoff = 0; |
| 1657 | loff_t size; |
| 1658 | |
Yan, Zheng | 494d77b | 2014-06-26 15:25:17 +0800 | [diff] [blame] | 1659 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
| 1660 | return -EOPNOTSUPP; |
| 1661 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1662 | if (!S_ISREG(inode->i_mode)) |
| 1663 | return -EOPNOTSUPP; |
| 1664 | |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1665 | prealloc_cf = ceph_alloc_cap_flush(); |
| 1666 | if (!prealloc_cf) |
| 1667 | return -ENOMEM; |
| 1668 | |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1669 | inode_lock(inode); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1670 | |
| 1671 | if (ceph_snap(inode) != CEPH_NOSNAP) { |
| 1672 | ret = -EROFS; |
| 1673 | goto unlock; |
| 1674 | } |
| 1675 | |
| 1676 | if (ceph_osdmap_flag(osdc->osdmap, CEPH_OSDMAP_FULL) && |
| 1677 | !(mode & FALLOC_FL_PUNCH_HOLE)) { |
| 1678 | ret = -ENOSPC; |
| 1679 | goto unlock; |
| 1680 | } |
| 1681 | |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1682 | if (ci->i_inline_version != CEPH_INLINE_NONE) { |
| 1683 | ret = ceph_uninline_data(file, NULL); |
| 1684 | if (ret < 0) |
| 1685 | goto unlock; |
| 1686 | } |
| 1687 | |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1688 | size = i_size_read(inode); |
| 1689 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
| 1690 | endoff = offset + length; |
| 1691 | |
| 1692 | if (fi->fmode & CEPH_FILE_MODE_LAZY) |
| 1693 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; |
| 1694 | else |
| 1695 | want = CEPH_CAP_FILE_BUFFER; |
| 1696 | |
Yan, Zheng | 3738daa | 2014-11-14 22:10:07 +0800 | [diff] [blame] | 1697 | 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] | 1698 | if (ret < 0) |
| 1699 | goto unlock; |
| 1700 | |
| 1701 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
| 1702 | if (offset < size) |
| 1703 | ceph_zero_pagecache_range(inode, offset, length); |
| 1704 | ret = ceph_zero_objects(inode, offset, length); |
| 1705 | } else if (endoff > size) { |
| 1706 | truncate_pagecache_range(inode, size, -1); |
| 1707 | if (ceph_inode_set_size(inode, endoff)) |
| 1708 | ceph_check_caps(ceph_inode(inode), |
| 1709 | CHECK_CAPS_AUTHONLY, NULL); |
| 1710 | } |
| 1711 | |
| 1712 | if (!ret) { |
| 1713 | spin_lock(&ci->i_ceph_lock); |
Yan, Zheng | 28127bd | 2014-11-14 22:38:29 +0800 | [diff] [blame] | 1714 | ci->i_inline_version = CEPH_INLINE_NONE; |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1715 | dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, |
| 1716 | &prealloc_cf); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1717 | spin_unlock(&ci->i_ceph_lock); |
| 1718 | if (dirty) |
| 1719 | __mark_inode_dirty(inode, dirty); |
| 1720 | } |
| 1721 | |
| 1722 | ceph_put_cap_refs(ci, got); |
| 1723 | unlock: |
Al Viro | 5955102 | 2016-01-22 15:40:57 -0500 | [diff] [blame] | 1724 | inode_unlock(inode); |
Yan, Zheng | f66fd9f | 2015-06-10 17:26:13 +0800 | [diff] [blame] | 1725 | ceph_free_cap_flush(prealloc_cf); |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1726 | return ret; |
| 1727 | } |
| 1728 | |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1729 | const struct file_operations ceph_file_fops = { |
| 1730 | .open = ceph_open, |
| 1731 | .release = ceph_release, |
| 1732 | .llseek = ceph_llseek, |
Al Viro | 3644424 | 2014-04-02 20:28:01 -0400 | [diff] [blame] | 1733 | .read_iter = ceph_read_iter, |
Al Viro | 4908b82 | 2014-04-03 23:09:01 -0400 | [diff] [blame] | 1734 | .write_iter = ceph_write_iter, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1735 | .mmap = ceph_mmap, |
| 1736 | .fsync = ceph_fsync, |
Greg Farnum | 40819f6 | 2010-08-02 15:34:23 -0700 | [diff] [blame] | 1737 | .lock = ceph_lock, |
| 1738 | .flock = ceph_flock, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1739 | .splice_read = generic_file_splice_read, |
Al Viro | 3551dd7 | 2014-04-05 04:40:12 -0400 | [diff] [blame] | 1740 | .splice_write = iter_file_splice_write, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1741 | .unlocked_ioctl = ceph_ioctl, |
| 1742 | .compat_ioctl = ceph_ioctl, |
Li Wang | ad7a60d | 2013-08-15 11:51:44 +0800 | [diff] [blame] | 1743 | .fallocate = ceph_fallocate, |
Sage Weil | 124e68e | 2009-10-06 11:31:08 -0700 | [diff] [blame] | 1744 | }; |
| 1745 | |